Yan Liu, Feng Zhang, Zhanyu Ma, Jun Xu, Jiuchong Gao, Jinghua Hao, Renqing He, Han Liu, Yangdong Deng. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Weitao Ma, Xiaocheng Feng, Lei Huang, Xiachong Feng, Zhanyu Ma, Jun Xu, Jiuchong Gao, Jinghua Hao, Renqing He, Bing Qin. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Tool-Integrated Reasoning (TIR) empowers large language models (LLMs) to tackle complex tasks by interleaving reasoning steps with external tool interactions.However, existing reinforcement learning methods typically rely on outcome-or trajectory-level rewards, assigning uniform advantages to all steps within a trajectory.This coarse-grained credit assignment fails to distinguish effective tool calls from redundant or erroneous ones, particularly in long-horizon multi-turn scenarios.To address this, we propose MatchTIR, a framework that introduces fine-grained supervision via bipartite matching-based turn-level reward assignment and dual-level advantage estimation.Specifically, we formulate credit assignment as a bipartite matching problem between predicted and ground-truth traces, utilizing two assignment strategies to derive dense turn-level rewards.Furthermore, to balance local step precision with global task success, we introduce a dual-level advantage estimation scheme that integrates turn-level and trajectory-level signals, assigning distinct advantage values to individual interaction turns.Extensive experiments on three benchmarks demonstrate the superiority of MatchTIR.Notably, our 4B model surpasses the majority of 8B competitors, particularly in long-horizon and multiturn tasks.Our codes are available at https: //github.com/quchangle1/MatchTIR.

Live streaming platforms have become a dominant form of online content consumption, offering dynamically evolving content, real-time interactions, and highly engaging user experiences. These unique characteristics introduce new challenges that differentiate live streaming recommendation from traditional recommendation settings and have garnered increasing attention from industry in recent years. However, research progress in academia has been hindered by the lack of publicly available datasets that accurately reflect the dynamic nature of live streaming environments. To address this gap, we introduce KuaiLive, the first real-time, interactive dataset collected from Kuaishou, a leading live streaming platform in China with over 400 million daily active users. The dataset records the interaction logs of 23,772 users and 452,621 streamers over a 21-day period. Compared to existing datasets, KuaiLive offers several advantages: it includes precise live room start and end timestamps, multiple types of real-time user interactions (click, comment, like, gift), and rich side information features for both users and streamers. These features enable more realistic simulation of dynamic candidate items and better modeling of user and streamer behaviors. We conduct a thorough analysis of KuaiLive from multiple perspectives and evaluate several representative recommendation methods on it, establishing a strong benchmark for future research. KuaiLive can support a wide range of tasks in the live streaming domain, such as top-K recommendation, click-through rate prediction, watch time prediction, and gift price prediction. Moreover, its fine-grained behavioral data also enables research on multi-behavior modeling, multi-task learning, and fairness-aware recommendation. We believe that KuaiLive will serve as a valuable resource to advance the development of intelligent live streaming services. The dataset and related resources are publicly available at: https://imgkkk574.github.io/KuaiLive.

Current benchmarks evaluate the performance of RAG methods from various perspectives, they share a common assumption that user queries used for retrieval are error-free. However, in real-world interactions between users and LLMs, query entry errors are frequent. The impact of these errors on current RAG methods against such errors remains largely unexplored. To bridge this gap, we propose QE-RAG, the first robust RAG benchmark designed specifically to evaluate performance against query entry errors. We analyze the impact of these errors on LLM outputs and find that corrupted queries degrade model performance, which can be mitigated through query correction and training a robust retriever for retrieving relevant documents. Based on these insights, we propose a contrastive learning-based robust retriever training method and a retrieval-augmented query correction method. Extensive experiments reveal that: (1) state-of-the-art RAG methods including sequential, branching, and iterative methods, exhibit poor robustness to query entry errors; (2) our method enhances the robustness of RAG when handling query entry errors and it's compatible with existing RAG methods, further improving their robustness.

As an emerging advertising pricing method, optimized cost-per-click (OCPC) has attracted much research interest. In OCPC, the platform employs intelligent bidding strategies to optimize the advertising performance. Although widely adopted, existing bidding strategies overlook the delayed conversion phenomenon in OCPC, that is, the platform needs to wait for a period to receive the corresponding conversion signal after a click. Ignoring such delayed conversions causes the bidding strategies to overestimate the cost-per-action (average cost of a conversion, CPA), bid low, and finally hurt the platform's revenue. Moreover, the characteristics of the OCPC scenario make estimating the conversion probabilities for the delayed conversions more difficult. To address these issues, this paper proposes SurvBid (bidding with Survival Analysis) which aims to predict the convert probabilities for those delayed conversions in OCPC scenario. The CPA can then be accurately estimated and used to guide existing bidding methods to make more accurate bids. To meet the needs of different advertising platforms, we provide two versions of SurvBid, SurvBid-M (SurvBid with multitask survival model) and SurvBid-C (SurvBid with Cox survival model) with theoretical results to guide the model selection. Both online and offline experiments demonstrate that SurvBid can improve the platform's revenue and advertisers' conversions.

Live streaming platforms have recently gained popularity due to their immediacy and entertainment value, highlighting the need for streaming recommender systems that can adapt to the dynamic nature of evolving content, real-time interactions, and changing user interests. The ''live room'' plays a central role in modeling this dynamic environment, as it not only connects users with streamers but also serves as a key channel for collecting fine-grained user feedback. More specifically, the frequent interactions of users within a live room provide detailed dynamic collaboration information, reflecting the streamer's real-time topic and users' dynamic interests. However, existing studies have not thoroughly investigated the dynamics of room-level collaborative information. In this paper, we address this gap by emphasizing two perspectives: the evolving tripartite interaction information among rooms, streamers, and users, and the real-time intra-room collaboration information. We propose DCGLive, a Dynamic Collaboration-aware Graph learning approach for Live streaming recommendation. Specifically, we first construct two dynamic bipartite graphs to perceive the evolving tripartite interaction and generate real-time representations of streamers, rooms, and users. To account for the dynamic nature of live streaming, we design a set of non-parametric, collaboration-aware indicators that weight intra-room interactions based on both temporal recency and frequency, while guiding the embedding updating process for both users and rooms. Additionally, to address the cold-start challenge of newly created live rooms in real time, we propose a room representation initialization mechanism that balances both the relevance and the dynamics among different rooms hosted by the same streamer. Experiments conducted on both commercial and public datasets demonstrate that DCGLive consistently outperforms the baseline models. Our code is available at https://github.com/imgkkk574/DCGLive.

Short-video applications have attracted substantial user traffic. However, these platforms also foster problematic usage patterns, commonly referred to as short-video addiction, which pose risks to both user health and the sustainable development of platforms. Prior studies on this issue have primarily relied on questionnaires or volunteer-based data collection, which are often limited by small sample sizes and population biases. In contrast, short-video platforms have large-scale behavioral data, offering a valuable foundation for analyzing addictive behaviors. To examine addiction-aware behavior patterns, we combine economic addiction theory with users' implicit behavior captured by recommendation systems. Our analysis shows that short-video addiction follows functional patterns similar to traditional forms of addictive behavior (e.g., substance abuse) and that its intensity is consistent with findings from previous social science studies. To develop a simulator that can learn and model these patterns, we introduce a novel training framework, AddictSim. To consider the personalized addiction patterns, AddictSim uses a mean-to-adapted strategy with group relative policy optimization training. Experiments on two large-scale datasets show that AddictSim consistently outperforms existing training strategies. Our simulation results show that integrating diversity-aware algorithms can mitigate addictive behaviors well.

Traditional imitation learning focuses on modeling the behavioral mechanisms of experts, which requires a large amount of interaction history generated by some fixed expert. However, in many streaming applications, such as streaming recommender systems, online decision-makers typically engage in online learning during the decision-making process, meaning that the interaction history generated by online decision-makers includes their behavioral evolution from novice expert to experienced expert. This poses a new challenge for existing imitation learning approaches that can only utilize data from experienced experts. To address this issue, this paper proposes an inverse batched contextual bandit (IBCB) framework that can efficiently perform estimations of environment reward parameters and learned policy based on the expert's behavioral evolution history. Specifically, IBCB formulates the inverse problem into a simple quadratic programming problem by utilizing the behavioral evolution history of the batched contextual bandit with inaccessible rewards, and it can be extended to fairness-aware expert limitation. We demonstrate that IBCB is a unified framework for both deterministic and randomized bandit policies. The experimental results indicate that IBCB outperforms several existing imitation learning algorithms on synthetic and real-world data and significantly reduces running time. Additionally, empirical analyses reveal that IBCB exhibits better imitation ability for fairness-aware experts, out-of-distribution generalization and is highly effective in learning the bandit policy from the interaction history of novice experts. The code is publicly available.

Sequential Recommendation (SeqRec) aims to predict the next item by capturing sequential patterns from users' historical interactions, playing a crucial role in many real-world recommender systems. However, existing approaches predominantly adopt a direct forward computation paradigm, where the final hidden state of the sequence encoder serves as the user representation. We argue that this inference paradigm, due to its limited computational depth, struggles to model the complex evolving nature of user preferences and lacks a nuanced understanding of long-tail items, leading to suboptimal performance. To address this issue, we propose ReaRec, the first inference-time computing framework for recommender systems, which enhances user representations through implicit multi step reasoning. Specifically, ReaRecautoregressively feeds the sequence's last hidden state into the sequential recommender while incorporating special reasoning position embeddings to decouple the original item encoding space from the multi step reasoning space. Moreover, we introduce two lightweight reasoning-based learning methods, Ensemble Reasoning Learning (ERL) and Progressive Reasoning Learning (PRL), to further effectively exploit ReaRec's reasoning potential. Extensive experiments on five public real-world datasets and different SeqRec architectures demonstrate the generality and effectiveness of our proposed ReaRec. Remarkably, post-hoc analyses reveal that ReaRec significantly elevates the performance ceiling of multiple sequential recommendation backbones by approximately 30% 50%. Thus, we believe this work can open a new and promising avenue for future research in inference-time computing for sequential recommendation. The source code is available at https://github.com/TangJiakai/ReaRec.git.

In real-world streaming recommender systems, user preferences evolve dynamically over time. Existing bandit-based methods treat time merely as a timestamp, neglecting its explicit relationship with user preferences and leading to suboptimal performance. Moreover, the online learning methods often suffer from inefficient exploration–exploitation during the early online phase. To address these issues, we propose HyperBandit+, a novel contextual bandit policy which integrates a time-aware hypernetwork to adapt to time-varying user preferences and employs a large language model-assisted warm-start mechanism (LLM Start) to enhance exploration–exploitation efficiency at the early online phase. Specifically, HyperBandit+ leverages a neural network that takes time features as input and generates parameters for estimating time-varying rewards by capturing the correlation between time and user preferences. Additionally, the LLM Start mechanism employs multi-step data augmentation to simulate realistic interaction data for effective offline learning, providing warm-start parameters for the bandit policy at the early online phase. To meet real-time streaming recommendation demands, we adopt low-rank factorization to reduce hypernetwork training complexity. Theoretically, we rigorously establish a sublinear regret upper bound that accounts for both the hypernetwork and the LLM warm-start mechanism. Extensive experiments on real-world datasets demonstrate that HyperBandit+ consistently outperforms state-of-the-art baselines in terms of accumulated rewards.

In multi-stakeholder recommender systems (RS), users and providers operate as two crucial and interdependent roles, whose interests must be well-balanced. Prior research has demonstrated the importance of guaranteeing both provider fairness and user accuracy to meet their interests. However, when balancing the two objectives, another critical factor emerges: individual fairness, which manifests as a significant disparity in individual recommendation accuracy, with some users receiving high accuracy while others are left with notably low accuracy. This oversight severely harms the interests of users and exacerbates social polarization. How to guarantee individual fairness while ensuring user accuracy and provider fairness remains an unsolved problem. To bridge this gap, this article proposes a method called BankFair+, which extends BankFair with two steps: (1) introducing a non-linear function from regret theory to ensure individual fairness while enhancing user accuracy; (2) formulating the re-ranking process as a regret-aware fuzzy programming problem to meet the interests of both individual users and providers, therefore balancing the tradeoff between individual fairness and provider fairness. Experiments on two real-world recommendation datasets demonstrate that BankFair+ outperforms all baselines regarding individual fairness, user accuracy, and provider fairness, indicating its ability to guarantee two-sided fairness and accuracy in RS.

Online to batch conversion involves constructing a new batch learner by utilizing a series of models generated by an existing online learning algorithm, for achieving generalization guarantees under i.i.d assumption. However, when applied to real-world streaming applications such as streaming recommender systems, the data stream may be sampled from time-varying distributions instead of persistently being i.i.d. This poses a challenge in terms of out-of-distribution (OOD) generalization. Existing approaches employ fixed conversion mechanisms that are unable to adapt to novel testing distributions, hindering the testing accuracy of the batch learner. To address these issues, we propose AdaO2B, an adaptive online to batch conversion approach under the bandit setting. AdaO2B is designed to be aware of the distribution shifts in the testing data and achieves OOD generalization guarantees. Specifically, AdaO2B can dynamically combine the sequence of models learned by a contextual bandit algorithm and determine appropriate combination weights using a context-aware weighting function. This innovative approach allows for the conversion of a sequence of models into a batch learner that facilitates OOD generalization. Theoretical analysis provides justification for why and how the learned adaptive batch learner can achieve OOD generalization error guarantees. Experimental results have demonstrated that AdaO2B significantly outperforms state-of-the-art baselines on both synthetic and real-world recommendation datasets.

In search scenarios, user experience can be hindered by erroneous queries due to typos, voice errors, or knowledge gaps. Therefore, query correction is crucial for search engines. Current correction models, usually small models trained on specific data, often struggle with queries beyond their training scope or those requiring contextual understanding. While the advent of Large Language Models (LLMs) offers a potential solution, they are still limited by their pre-training data and inference cost, particularly for complex queries, making them not always effective for query correction. To tackle these, we propose Trigger3, a large-small model collaboration framework that integrates the traditional correction model and LLM for query correction, capable of adaptively choosing the appropriate correction method based on the query and the correction results from the traditional correction model and LLM. Trigger3 first employs a correction trigger to filter out correct queries. Incorrect queries are then corrected by the traditional correction model. If this fails, an LLM trigger is activated to call the LLM for correction. Finally, for queries that no model can correct, a fallback trigger decides to return the original query. Extensive experiments demonstrate Trigger3 outperforms correction baselines while maintaining efficiency.

Continual pre-training (CPT) has been an important approach for adapting language models to specific domains or tasks. In this paper, we comprehensively study its key designs to balance the new abilities while retaining the original abilities, and present an effective CPT method that can greatly improve the Chinese language ability and scientific reasoning ability of LLMs. To achieve it, we design specific data mixture and curriculum strategies based on existing datasets and synthetic high-quality data. Concretely, we synthesize multidisciplinary scientific QA pairs based on related web pages to guarantee the data quality, and also devise the performance tracking and data mixture adjustment strategy to ensure the training stability. For the detailed designs, we conduct preliminary studies on a relatively small model, and summarize the findings to help optimize our CPT method. Extensive experiments on a number of evaluation benchmarks show that our approach can largely improve the performance of Llama-3 (8B), including both the general abilities (+8.81 on C-Eval and +6.31 on CMMLU) and the scientific reasoning abilities (+12.00 on MATH and +4.13 on SciEval). Our model, data, and codes are available at https://github.com/RUC-GSAI/Llama-3-SynE.

Weicong Qin, Yi Xu, Weijie Yu, Chenglei Shen, Ming He, Jianping Fan, Xiao Zhang, Jun Xu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Text embeddings from PLM-based models enable a wide range of applications, yet their performance often degrades on longer texts.In this paper, we introduce a phenomenon we call Length Collapse, where embeddings of longer texts tend to cluster together.This clustering results in a distributional inconsistency between the embeddings of short and long texts.We further investigate how these differences contribute to the performance decline observed with longer texts across various downstream tasks.Through a rigorous theoretical analysis of the self-attention mechanism, which acts as a low-pass filter in PLM-based models, we demonstrate that as text length increases, the strength of low-pass filtering intensifies, causing embeddings to retain more low-frequency components.As a result, input token features become more similar, leading to clustering and ultimately the collapse of embeddings for longer texts.To address this issue, we propose a simple method, TempScale, which mitigates the Length Collapse phenomenon.By narrowing the gap in low-pass filtering rates between long and short texts, TempScale ensures more consistent embeddings across different text lengths.This approach leads to performance improvements of 0.94% on MTEB and 1.10% on LongEmbed, which focuses specifically on long-context retrieval, providing strong evidence for the validity of our analysis.

Recently, research found that the documents retrieved from the Retrieval-Augmented Generation (RAG) may contain conflicting knowledge with each other, leading Large Language Models (LLMs) to generate incorrect responses. To solve such a problem, existing approaches usually only keep the most frequently mentioned knowledge from these documents, since they assume that the most representative knowledge aligns best with the true answer. Although effective in certain scenarios, these approaches often underperform when the most frequent knowledge is not the correct one. From the voting perspective, these methods can be regarded as a Majority Voting (MV) process, which chooses the most frequent candidates among different candidate knowledge. However, we show that the underperformance of such methods stems from that MV is only effective with a small number of candidates and binary voting scores. In contrast, in the RAG scenario, the candidates (knowledge) are very diverse, and the voting scores (document relevance scores) are typically continuous. Simply adapting MV in RAG will result in poor performance of LLMs. In voting theory, on the other hand, the preference-based voting methods represented by the Borda Voting (BV) consider the whole preference order of voters over all candidates, enabling the selection of candidates that better represent the collective viewpoint. Inspired by such an insight, we propose BordaRAG, a model designed to better select the most appropriate documents from conflicting documents. Specifically, BordaRAG first computes the preference scores of the documents over the candidate answers. After that, a BV component is designed to select the winning documents according to the preference scores. Finally, the chosen documents are provided to LLMs, which will generate the final response. Experimental results on three open-domain QA datasets show that BordaRAG can outperform all baselines.

In modern online platforms, search and recommendation (S&R) often coexist, offering opportunities for performance improvement through search-enhanced approaches. Existing studies show that incorporating search signals boosts recommendation performance. However, the effectiveness of these methods relies heavily on rich search interactions. They primarily benefit a small subset of users with abundant search behavior, while offering limited improvements for the majority of users who exhibit only sparse search activity. To address the problem of sparse search data in search-enhanced recommendation, we face two key challenges : (1) how to learn useful search features for users with sparse search interactions, and (2) how to design effective training objectives under sparse conditions. Our idea is to leverage the features of users with rich search interactions to enhance those of users with sparse search interactions. Based on this idea, we propose GSERec, a method that utilizes message passing on the User-Code Graphs to alleviate data sparsity in Search-Enhanced Recommendation. Specifically, we utilize Large Language Models (LLMs) with vector quantization to generate discrete codes, which connect similar users and thereby construct the graph. Through message passing on this graph, embeddings of users with rich search data are propagated to enhance the embeddings of users with sparse interactions. To further ensure that the message passing captures meaningful information from truly similar users, we introduce a contrastive loss to better model user similarities. The enhanced user representations are then integrated into downstream search-enhanced recommendation models. Experiments on three real-world datasets show that GSERec consistently outperforms baselines, especially for users with sparse search behaviors.

Syllogistic reasoning is a fundamental aspect of legal decision-making, enabling logical conclusions by connecting general legal principles with specific case facts. Although existing large language models (LLMs) can generate responses to legal questions, they fail to perform explicit syllogistic reasoning, often producing implicit and unstructured answers that lack explainability and trustworthiness. To address this limitation, we propose SyLeR, a novel framework that empowers LLMs to engage in explicit syllogistic legal reasoning. SyLeR integrates a tree-structured hierarchical retrieval mechanism to effectively combine relevant legal statutes and precedent cases, forming comprehensive major premises. This is followed by a two-stage fine-tuning process: supervised fine-tuning warm-up establishes a foundational understanding of syllogistic reasoning, while reinforcement learning with a structure-aware reward mechanism refines the model's ability to generate diverse logically sound and well-structured reasoning paths. We conducted extensive experiments across various dimensions, including in-domain and cross-domain user groups (legal laypersons and practitioners), multiple languages (Chinese and French), and different LLM backbones (legal-specific and open-domain LLMs). The results show that SyLeR significantly improves response accuracy and consistently delivers explicit, explainable, and trustworthy legal reasoning.

Personalized retrieval-augmented generation (RAG) aims to produce user-tailored responses by incorporating retrieved user profiles alongside the input query. Existing methods primarily focus on improving retrieval and rely on large language models (LLMs) to implicitly integrate the retrieved context with the query. However, such models are often sensitive to retrieval quality and may generate responses that are misaligned with user preferences. To address this limitation, we propose PrLM, a reinforcement learning framework that trains LLMs to explicitly reason over retrieved user profiles. Guided by a contrastively trained personalization reward model, PrLM effectively learns from user responses without requiring annotated reasoning paths. Experiments on three personalized text generation datasets show that PrLM outperforms existing methods and remains robust across varying numbers of retrieved profiles and different retrievers.

Unlike traditional search engines that present ranked lists of webpages, generative search engines rely solely on in-line citations as the key gateway to original real-world webpages, making it crucial to examine whether LLMgenerated citations have biases-particularly for politically sensitive queries.To investigate this, we first construct AllSides-2024, a new dataset comprising the latest real-world news articles (Jan.2024 -Dec.2024) labeled with left-or right-leaning stances.Through systematic evaluations, we find that LLMs exhibit a consistent tendency to cite left-leaning sources at notably higher rates compared to traditional retrieval systems (e.g., BM25 and dense retrievers).Controlled experiments further reveal that this bias arises from a preference for media outlets identified as left-leaning, rather than for left-oriented content itself.Meanwhile, our findings show that while LLMs struggle to infer political bias from news content alone, they can almost perfectly recognize the political orientation of media outlets based on their names.These insights highlight the risk that, in the era of generative search engines, information exposure may be disproportionately shaped by specific media outlets, potentially shaping public perception and decision-making 1 .

Personalized search systems in e-commerce platforms increasingly involve user interactions with AI assistants, where users consult about products, usage scenarios, and more.Leveraging consultation to personalize search services is trending.Existing methods typically rely on semantic similarity to align historical consultations with current queries due to the absence of 'value' labels, but we observe that semantic similarity alone often fails to capture the true value of consultation for personalization.To address this, we propose a consultation value assessment framework that evaluates historical consultations from three novel perspectives: (1) Scenario Scope Value, (2) Posterior Action Value, and (3) Time Decay Value.Based on this, we introduce VAPS, a value-aware personalized search model that selectively incorporates high-value consultations through a consultation-user action interaction module and an explicit objective that aligns consultations with user actions.Experiments on both public and commercial datasets show that VAPS consistently outperforms baselines in both retrieval and ranking tasks.Codes are available at https://github.com/E-qin/VAPS.

Group max-min fairness (MMF) is commonly used in fairness-aware recommender systems (RS) as an optimization objective, as it aims to protect marginalized item groups and ensures a fair competition platform. However, our theoretical analysis indicates that integrating MMF constraint violates the assumption of sample independence during optimization, causing the loss function to deviate from linear additivity. Such nonlinearity property introduces the Jensen gap between the model's convergence point and the optimal point if mini-batch sampling is applied. Both theoretical and empirical studies show that as the mini-batch size decreases and the group size increases, the Jensen gap will widen accordingly. Some methods using heuristic re-weighting or debiasing strategies have the potential to bridge the Jensen gap. However, they either lack theoretical guarantees or suffer from heavy computational costs. To overcome these limitations, we first theoretically demonstrate that the MMF-constrained objective can be essentially reformulated as a group-weighted optimization objective. Then we present an efficient and effective algorithm named FairDual, which utilizes a dual optimization technique to minimize Jensen gap. Our theoretical analysis demonstrates that FairDual can achieve a sub-linear convergence rate to the globally optimal solution and the Jensen gap can be well bounded under a mini-batch sampling strategy with random shuffle. Extensive experiments conducted using six large-scale RS backbone models on three publicly available datasets demonstrate that FairDual outperforms all baselines in terms of both accuracy and fairness.

Tool learning enables Large Language Models (LLMs) to interact with external environments by invoking tools, serving as an effective strategy to mitigate the limitations inherent in their pre-training data. In this process, tool documentation plays a crucial role by providing usage instructions for LLMs, thereby facilitating effective tool utilization. This paper concentrates on the critical challenge of bridging the comprehension gap between LLMs and external tools due to the inadequacies and inaccuracies inherent in existing human-centric tool documentation. We propose a novel framework, DRAFT, aimed at Dynamically Refining tool documentation through the Analysis of Feedback and Trials emanating from LLMs' interactions with external tools. This methodology pivots on an innovative trial-and-error approach, consisting of three distinct learning phases: experience gathering, learning from experience, and documentation rewriting, to iteratively enhance the tool documentation. This process is further optimized by implementing a diversity-promoting exploration strategy to ensure explorative diversity and a tool-adaptive termination mechanism to prevent overfitting while enhancing efficiency. Extensive experiments on multiple datasets demonstrate that DRAFT's iterative, feedback-based refinement significantly ameliorates documentation quality, fostering a deeper comprehension and more effective utilization of tools by LLMs. Notably, our analysis reveals that the tool documentation refined via our approach demonstrates robust cross-model generalization capabilities.

Retrieval-Augmented Generation (RAG) models are designed to incorporate external knowledge, reducing hallucinations caused by insufficient parametric (internal) knowledge. However, even with accurate and relevant retrieved content, RAG models can still produce hallucinations by generating outputs that conflict with the retrieved information. Detecting such hallucinations requires disentangling how Large Language Models (LLMs) balance external and parametric knowledge. Current detection methods often focus on one of these mechanisms or without decoupling their intertwined effects, making accurate detection difficult. In this paper, we investigate the internal mechanisms behind hallucinations in RAG scenarios. We discover hallucinations occur when the **Knowledge FFNs** in LLMs overemphasize parametric knowledge in the residual stream, while **Copying Heads** fail to effectively retain or integrate external knowledge from retrieved content. Based on these findings, we propose **ReDeEP**, a novel method that detects hallucinations by decoupling LLM’s utilization of external context and parametric knowledge. Our experiments show that ReDeEP significantly improves RAG hallucination detection accuracy. Additionally, we introduce AARF, which mitigates hallucinations by modulating the contributions of Knowledge FFNs and Copying Heads.

Previous studies have found that PLM-based retrieval models exhibit a preference for LLM-generated content, assigning higher relevance scores to these documents even when their semantic quality is comparable to human-written ones. This phenomenon, known as source bias, threatens the sustainable development of the information access ecosystem. However, the underlying causes of source bias remain unexplored. In this paper, we explain the process of information retrieval with a causal graph and discover that PLM-based retrievers learn perplexity features for relevance estimation, causing source bias by ranking the documents with low perplexity higher. Theoretical analysis further reveals that the phenomenon stems from the positive correlation between the gradients of the loss functions in language modeling task and retrieval task. Based on the analysis, a causal-inspired inference-time debiasing method is proposed, called **C**ausal **D**iagnosis and **C**orrection (CDC). CDC first diagnoses the bias effect of the perplexity and then separates the bias effect from the overall estimated relevance score. Experimental results across three domains demonstrate the superior debiasing effectiveness of CDC, emphasizing the validity of our proposed explanatory framework. Source codes are available at https://github.com/WhyDwelledOnAi/Perplexity-Trap.

Recent Graphical User Interface (GUI) agents replicate the R1-Zero paradigm, coupling online Reinforcement Learning (RL) with explicit chain-of-thought reasoning prior to object grounding and thereby achieving substantial performance gains. In this paper, we first conduct extensive analysis experiments of three key components of that training pipeline: input design, output evaluation, and policy update—each revealing distinct challenges arising from blindly applying general-purpose RL without adapting to GUI grounding tasks. Input design: Current templates encourage the model to generate chain-of-thought reasoning, but longer chains unexpectedly lead to worse grounding performance. Output evaluation: Reward functions based on hit signals or box area allow models to exploit box size, leading to reward hacking and poor localization quality. Policy update: Online RL tends to overfit easy examples due to biases in length and sample difficulty, leading to under-optimization on harder cases. To address these issues, we propose three targeted solutions. First, we adopt a $\textbf{Fast Thinking Template}$ that encourages direct answer generation, reducing excessive reasoning during training. Second, we incorporate a box size constraint into the reward function to mitigate reward hacking. Third, we revise the RL objective by adjusting length normalization and adding a difficulty-aware scaling factor, enabling better optimization on hard samples. Our $\textbf{GUI-G1-3B}$, trained on 17K public samples with Qwen2.5-VL-3B-Instruct, achieves $\textbf{90.3\%}$ accuracy on ScreenSpot and $\textbf{37.1\%}$ on ScreenSpot-Pro. This surpasses all prior models of similar size and even outperforms the larger UI-TARS-7B, establishing a new state-of-the-art in GUI agent grounding.

Recommender systems are essential for information access, allowing users to present their content for recommendation. With the rise of large language models (LLMs), AI-generated content (AIGC), primarily in the form of text, has become a central part of the content ecosystem. As AIGC becomes increasingly prevalent, it is important to understand how it affects the performance and dynamics of recommender systems. To this end, we construct an environment that incorporates AIGC to explore its short-term impact. The results from popular sequential recommendation models reveal that AIGC are ranked higher in the recommender system, reflecting the phenomenon of source. To further explore the long-term impact of AIGC, we introduce a feedback loop with realistic simulators. The results show that the model's preference for AIGC increases as the user clicks on AIGC rises and the model trains on simulated click data. This leads to two issues: In the short term, bias toward AIGC encourages LLM-based content creation, increasing AIGC content, and causing unfair traffic distribution. From a long-term perspective, our experiments also show that when AIGC dominates the content ecosystem after a feedback loop, it can lead to a decline in recommendation performance. To address these issues, we propose a debiasing method based on L1-loss optimization to maintain long-term content ecosystem balance. In a real-world environment with AIGC generated by mainstream LLMs, our method ensures a balance between AIGC and human-generated content in the ecosystem. The code and dataset are available at https://github.com/Yuqi-Zhou/Rec_SourceBias.

In modern information retrieval (IR), going beyond accuracy is crucial for maintaining a healthy ecosystem, particularly in meeting fairness and diversity requirements. To address these needs, various datasets, algorithms, and evaluation methods have been developed. These algorithms are often tested with different metrics, datasets, and experimental settings, making comparisons inconsistent and challenging. Consequently, there is an urgent need for a comprehensive IR toolkit, enabling standardized assessments of fairness- and diversity-aware algorithms across IR tasks. To address these issues, we introduce an open-source standardized toolkit called FairDiverse. First, FairDiverse provides a comprehensive framework for incorporating fairness- and diversity-aware approaches, including pre-processing, in-processing, and post-processing methods, into different pipeline stages of IR. Second, FairDiverse enables the evaluation of 29 fairness, and diversity algorithms across 16 base models for two fundamental IR tasks-search and recommendation-facilitating the establishment of a comprehensive benchmark. Finally, FairDiverse is highly extensible, offering multiple APIs to enable IR researchers to quickly develop their own fairness- and diversity-aware IR models, and allows for fair comparisons with existing baselines. The project is open-sourced on GitHub:~ https://github.com/XuChen0427/FairDiverse.

Fairness-aware information retrieval (IR) has attracted growing attention, with numerous metrics and algorithms proposed. However, the complexity of fairness and IR systems makes it challenging to systematically summarize progress and identify future directions. Economics has long studied fairness and offers a system-oriented perspective that naturally captures societal and intertemporal trade-offs. In this tutorial, we first frame IR systems as specialized economic markets and reorganize fairness algorithms along three key economic dimensions: macro vs. micro, demand vs. supply, and short-term vs. long-term. Unlike prior fairness-aware tutorials, this economic lens not only provides a structured reframing of fairness-aware IR, but also points toward new opportunities by encouraging the use of economic tools to address open problems.

Fairness is an increasingly important factor in re-ranking tasks. Prior work has identified a trade-off between ranking accuracy and item fairness. However, the underlying mechanisms are still not fully understood. An analogy can be drawn between re-ranking and the dynamics of economic transactions. The accuracy-fairness trade-off parallels the coupling of the commodity tax transfer process. Fairness considerations in re-ranking, similar to a commodity tax on suppliers, ultimately translate into a cost passed on to consumers. Analogously, item-side fairness constraints result in a decline in user-side accuracy. In economics, the extent to which commodity tax on the supplier (item fairness) transfers to commodity tax on users (accuracy loss) is formalized using the notion of elasticity. The re-ranking fairness-accuracy trade-off is similarly governed by the elasticity of utility between item groups. This insight underscores the limitations of current fair re-ranking evaluations, which often rely solely on a single fairness metric, hindering comprehensive assessment of fair re-ranking algorithms.

Recent studies have revealed a phenomenon known as source bias, where PLM-based retrievers assign higher relevance scores to LLM-generated content despite its semantic quality being comparable to human-written content. As LLMs rapidly advance and become more widely used, effectively counteracting source bias is crucial for the sustainable development of the information retrieval (IR) ecosystem. Existing methods primarily attempt to address source bias from the retriever side, adopting a "passive defense" approach that intervenes only after biased content has entered the retrieval pipeline. These solutions are limited by frequent retriever updates in industrial applications, high recurring costs, and their inability to address the root cause of source bias.

Generative AI search driven by large language models (LLMs) is reshaping information retrieval by offering end-to-end answers to complex queries, reducing users' reliance on manually browsing and summarizing multiple web pages. However, while this paradigm enhances convenience, it disrupts the feedback-driven improvement loop that has historically powered the evolution of traditional Web search. Web search can continuously improve their ranking models by collecting large-scale, fine-grained user feedback (e.g., clicks, dwell time) at the document level. In contrast, generative AI search operates through a much longer search pipeline-spanning query decomposition, document retrieval, and answer generation-yet typically receives only coarse-grained feedback on the final answer. This introduces a feedback loop disconnect, where user feedback for the final output cannot be effectively mapped back to specific system components, making it difficult to improve each intermediate stage and sustain the feedback loop.

Recently, live streaming services have seen a surge in popularity, prompting many platforms to offer both short video and live streaming services to meet the diverse needs of users and streamers. This has resulted in a close connection between short videos and live streaming within these platforms. Incorporating short video data into live streaming recommendation through cross-domain approaches can effectively mitigate the sparsity of live streaming gifting data. However, existing cross-domain recommendation methods primarily focus on transferring information across domains through overlapping users or items, while overlooking the strong connection between non-overlapping short videos and streamers. In this paper, we propose MGCCDR, a Multi-Graph Contrastive learning framework for Cross-Domain Recommendation, which leverages both overlapping users and non-overlapping items to enhance information transfer. Specifically, we first learn global representations from a global graph to establish connections between streamers and short videos. Subsequently, we construct three bipartite graphs among users, authors, and videos and introduce multi-graph learning to capture preferences within the target domain view, the source domain view, and the cross-domain view. Additionally, to address the varying contributions of each graph to the final recommendation task, we design an attention-based method to effectively integrate these representations, facilitating the information aggregation across domains. Extensive experiments on both commercial and public datasets demonstrate that our MGCCDR significantly outperforms the state-of-the-art methods.

Recently, the personalization of Large Language Models (LLMs) to generate content that aligns with individual user preferences has garnered widespread attention. Personalized Retrieval-Augmented Generation (RAG), which retrieves relevant documents from the user's history to reflect their preferences and enhance LLM generation, is one commonly used approach for personalization. However, existing personalized RAG methods do not consider that the histories of similar users can also assist in personalized generation for the current user, meaning that collaborative information between users can also benefit personalized generation. Inspired by the application of collaborative filtering in recommender systems, we propose a method called CFRAG, which adapts Collaborative Filtering to RAG for personalized text generation. However, this presents two challenges: (1) how to incorporate collaborative information without explicit user similarity labels? (2) how to retrieve documents that support personalized LLM generation? For Challenge 1, we use contrastive learning to train user embeddings to retrieve similar users and introduce collaborative information. For Challenge 2, we design a personalized retriever and reranker to retrieve the top-k documents from these users' histories. We take into account the user's preference during retrieval and reranking. Then we leverage feedback from the LLM to fine-tune the personalized retriever and reranker, enabling them to retrieve documents that meet the personalized generation needs of the LLM. Experimental results on the Language Model Personalization (LaMP) benchmark validate the effectiveness of CFRAG. Further analysis confirms the importance of incorporating collaborative information.

Retrieval-Augmented Generation (RAG) systems for Large Language Models (LLMs) have shown promise in knowledge-intensive tasks, yet their reasoning capabilities, particularly for complex multi-step reasoning, remain limited. Although recent approaches have explored integrating RAG with chain-of-thought reasoning or incorporating test-time search with process reward model (PRM), these methods face several untrustworthy challenges, including lack of explanations, bias in PRM training data, early-step bias in PRM scores, and ignoring post-training that fails to fully optimize reasoning potential. To address these issues, we propose Retrieval-Augmented Reasoning through Trustworthy Process Rewarding (ReARTeR), a framework that enhances RAG systems' reasoning capabilities through both post-training and test-time scaling. At test time, ReARTeR introduces Trustworthy Process Rewarding via a Process Reward Model for accurate scalar scoring and a Process Explanation Model (PEM) for generating natural language explanations, enabling step refinement. During post-training, we leverage Monte Carlo Tree Search guided by Trustworthy Process Rewarding to collect high-quality step-level preference data, which is used to optimize the model through Iterative Preference Optimization. ReARTeR tackles three key challenges: (1) misalignment between PRM and PEM, addressed through off-policy preference learning; (2) bias in PRM training data, mitigated by a balanced annotation method and incorporating stronger annotations for difficult examples; and (3) early-step bias in PRM, resolved via a temporal-difference-based look-ahead search strategy. Experimental results on multi-step reasoning benchmarks demonstrate that ReARTeR significantly improves reasoning performance, highlighting its potential to advance the reasoning capability of RAG systems.

Maintaining the long-term sustainability of recommender systems (RS) is crucial.Traditional RS evaluation methods primarily focus on the user's immediate feedback (e.g., click), however, they often overlook the long-term effect involved by the content creators.In the real world, content creators can strategically create and upload new items to the platform by analyzing users' feedback and preference trends.Although previous studies have attempted to model creator behaviors, they often overlook that such behaviors are under conditions of information asymmetry.This asymmetry arises because creators mainly access the user feedback on the items they produce, while the platform has access to the full spectrum of feedback data.However, existing RS simulators often fail to consider such a condition, making the long-term RS evaluation inaccurate.To bridge this gap, we propose a Large Language Model (LLM)empowered creator simulation agent named CreAgent.By utilizing the belief mechanism from game theory and the fast-and-slow thinking framework, we can simulate the creator's behaviors well under information asymmetry.Furthermore, to enhance CreAgent's simulation ability, we utilize Proximal Policy Optimization to fine-tune CreAgent.Our credibility validation experiments demonstrate that our simulation environment effectively aligns with the behaviors of real-world platforms and creators, thereby enhancing the reliability of long-term evaluations in RS.Furthermore, leveraging this simulator, we can examine whether RS algorithms, such as fairnessand diversity-aware methods, contribute to improving long-term performance for different stakeholders.

With the rapid advancements of large language models (LLMs), information retrieval (IR) systems, such as search engines and recommender systems, have undergone a paradigm shift due to their integration. However, integrating LLMs into the IR pipelines has also introduced new challenges, particularly in the form of biases and unfairness that may disrupt the information ecosystem. This tutorial will offer a comprehensive overview of emerging and pressing bias and unfairness issues associated with integrating LLMs into IR systems. Specifically, this tutorial first unifies bias and unfairness issues as problems of distribution mismatch and further categorizes the mitigation strategies under the umbrella of distribution alignment. Then, we summarize several types of bias and unfairness issues emerging from three critical stages of LLM integration into IR systems: data collection, model development, and result evaluation. We will systematically review and analyze their definitions, characteristics, and corresponding mitigation strategies in recent literature. Finally, we will highlight some open problems and future research directions. We hope this tutorial can raise the awareness of researchers and stakeholders in the IR field and beyond regarding bias and unfairness issues in this LLM era.

Recent research on query generation has focused on using Large Language Models (LLMs), which, despite achieving state-of-the-art performance, also introduce hallucination issues in generated queries. In this work, we categorize these issues into relevance hallucination and factuality hallucination, proposing a new typology for hallucinations arising from LLM-based query generation. We present an effective approach to decouple content from form in LLM-generated queries, preserving the factual knowledge extracted and integrated from inputs while leveraging the LLM's linguistic capabilities to construct syntactic structures, including function words. Specifically, we introduce a model-agnostic and training-free method that transforms the Large Language Model into a Pointer-Generator (LargePiG), where the pointer attention distribution utilizes the LLM's inherent attention weights, and the copy probability is derived from the difference between the vocabulary distribution in the model's high layers and the last layer. To validate the effectiveness of LargePiG, we constructed two datasets for assessing hallucination issues in query generation, covering both document and video scenarios. Empirical studies on various LLMs demonstrated LargePiG's superiority across both datasets. Additional experiments further verified that LargePiG reduces hallucination in large vision-language models and enhances the accuracy of document-based question-answering and factuality evaluation tasks. The source code and dataset are available at https://github.com/Jeryi-Sun/LargePiG.

Simulating high quality user behavior data has always been a fundamental yet challenging problem in human-centered applications such as recommendation systems, social networks, among many others. The major difficulty of user behavior simulation originates from the intricate mechanism of human cognitive and decision processes. Recently, substantial evidence has suggested that by learning huge amounts of web knowledge, large language models (LLMs) can achieve human-like intelligence and generalization capabilities. Inspired by such capabilities, in this article, we take an initial step to study the potential of using LLMs for user behavior simulation in the recommendation domain. To make LLMs act like humans, we design profile, memory and action modules to equip them, building LLM-based agents to simulate real users. To enable interactions between different agents and observe their behavior patterns, we design a sandbox environment, where each agent can interact with the recommendation system, and different agents can converse with their friends via one-to-one chatting or one-to-many social broadcasting. In the experiments, we first demonstrate the believability of the agent-generated behaviors based on both subjective and objective evaluations. Then, to show the potential applications of our method, we simulate and study two social phenomena including (1) information cocoons and (2) user conformity behaviors. We find that controlling the personalization degree of recommendation algorithms and improving the heterogeneity of user social relations can be two effective strategies for alleviating the problem of information cocoon, and the conformity behaviors can be highly influenced by the amount of user social relations. To advance this direction, we have released our project at https://github.com/RUC-GSAI/YuLan-Rec .

Multi-stakeholder recommender systems involve various roles, such as users and providers. Previous work pointed out that max-min fairness (MMF) is a better metric to support weak providers. However, when considering MMF, the features or parameters of these roles vary over time, and how to ensure long-term provider MMF has become a significant challenge. We observed that recommendation feedback loops (RFL) will influence the provider MMF greatly in the long term. RFL means that recommender systems can only receive feedback on exposed items from users and update recommender models incrementally based on this feedback. When utilizing the feedback, the recommender model will regard the unexposed items as negative. In this way, the tail provider will not get the opportunity to be exposed, and its items will always be considered negative samples. Such phenomena will become more and more serious in RFL. To alleviate the problem, this article proposes an online ranking model named Long-Term Provider Max-min Fairness (LTP-MMF). Theoretical analysis shows that the long-term regret of LTP-MMF enjoys a sub-linear bound. Experimental results on three public recommendation benchmarks demonstrated that LTP-MMF can outperform the baselines in the long term.

Nowadays, optimized cost-per-click (OCPC) has been widely adopted in online advertising. In OCPC, the advertiser sets an expected cost-per-conversion and pays per click, while the platform automatically adjusts the bid on each click to meet advertiser’s constraint. Existing bidding methods are based on feedback control, adjusting bids to keep the current cost-per-conversion close to the expected cost-per-conversion to avoid compensation. However, they overlook the conversion lag phenomenon: There always exists a time interval between the ad’s click time and conversion time. This interval makes existing methods overestimate the cost-per-conversion and results in over conservative bidding policies which finally hurts the revenue. To address the issue, this article proposes a novel bidding method, Bidding with Delayed Conversions (Bid-DC) which predicts the conversion probability of the clicked ads and used it to adjust the cost-per-conversion values. To ensure the bidding model can satisfy the advertiser’s constraint, constrained Markov decision process (CMDP) is adapted to automatically learn the optimal parameters from the log data. Both online and offline experiments demonstrate that Bid-DC outperforms the state-of-the-art baselines in terms of improving revenue. Empirical analysis also showed Bid-DC can accurately estimate the cost-per-conversion and make more stable bids.

Recently, integrating external tools with Large Language Models (LLMs) has gained significant attention as an effective strategy to mitigate the limitations inherent in their pre-training data. However, real-world systems often incorporate a wide array of tools, making it impractical to input all tools into LLMs due to length limitations and latency constraints. Therefore, to fully exploit the potential of tool-augmented LLMs, it is crucial to develop an effective tool retrieval system. Existing tool retrieval methods primarily focus on semantic matching between user queries and tool descriptions, frequently leading to the retrieval of redundant, similar tools. Consequently, these methods fail to provide a complete set of diverse tools necessary for addressing the multifaceted problems encountered by LLMs. In this paper, we propose a novel modelagnostic CO llaborative L earning-based T ool Retrieval approach, COLT, which captures not only the semantic similarities between user queries and tool descriptions but also takes into account the collaborative information of tools. Specifically, we first fine-tune the PLM-based retrieval models to capture the semantic relationships between queries and tools in the semantic learning stage. Subsequently, we construct three bipartite graphs among queries, scenes, and tools and introduce a dual-view graph collaborative learning framework to capture the intricate collaborative relationships among tools during the collaborative learning stage. Extensive experiments on both the open benchmark and the newly introduced ToolLens dataset show that COLT achieves superior performance. Notably, the performance of BERT-mini (11M) with our proposed model framework outperforms BERT-large (340M), which has 30 times more parameters. Furthermore, we will release ToolLens publicly to facilitate future research on tool retrieval.

Recent advancements in Large Language Models (LLMs) have attracted considerable interest among researchers to leverage these models to enhance Recommender Systems (RSs). Existing work predominantly utilizes LLMs to generate knowledge-rich texts or utilizes LLM-derived embeddings as features to improve RSs. Although the extensive world knowledge embedded in LLMs generally benefits RSs, the application can only take a limited number of users and items as inputs, without adequately exploiting collaborative filtering information. Considering its crucial role in RSs, one key challenge in enhancing RSs with LLMs lies in providing better collaborative filtering information through LLMs. In this paper, drawing inspiration from the in-context learning and chain of thought reasoning in LLMs, we propose the Large Language Models enhanced Collaborative Filtering (LLM-CF) framework, which distills the world knowledge and reasoning capabilities of LLMs into collaborative filtering. We also explored a concise and efficient instruction-tuning method, which improves the recommendation capabilities of LLMs while preserving their general functionalities (e.g., not decreasing on the LLM benchmark). Comprehensive experiments on three real-world datasets demonstrate that LLM-CF significantly enhances several backbone recommendation models and consistently outperforms competitive baselines, showcasing its effectiveness in distilling the world knowledge and reasoning capabilities of LLM into collaborative filtering.

Out of sustainable and economical considerations, two-sided recommendation platforms must satisfy the needs of both users and providers. Previous studies often show that the two sides' needs show different urgency: providers need a relatively long-term exposure demand while users want more short-term and accurate service. However, our empirical study reveals that previous methods for trading off fairness-accuracy often fail to guarantee long-term fairness and short-term accuracy simultaneously in real applications of fluctuating user traffic. Especially, when user traffic is low, the user experience often drops a lot. Our theoretical analysis also confirms that user traffic is a key factor in such a trade-off problem. How to guarantee accuracy and fairness under fluctuating user traffic remains a problem. Inspired by the bankruptcy problem in economics, we propose a novel fairness-aware re-ranking approach named BankFair. Intuitively, BankFair employs the Talmud rule to leverage periods of abundant user traffic to offset periods of user traffic scarcity, ensuring consistent user service at every period while upholding long-term fairness. Specifically, BankFair consists of two modules: (1) employing the Talmud rule to determine the required fairness degree under varying periods of user traffic; and (2) conducting an online re-ranking algorithm based on the fairness degree determined by the Talmud rule. Experiments on two real-world recommendation datasets show that BankFair outperforms all baselines regarding accuracy and provider fairness.

In real-world recommender systems, user engagement and subjective feedback play pivotal roles in shaping the content distribution mechanism of the platform. When platforms reach a certain scale, they often gather valuable questionnaire feedback data from users to evaluate their satisfaction with recommended items. Compared to traditional user feedback such as likes, questionnaires explicitly capture both satisfaction and dissatisfaction and are unaffected by other users' questionnaires, thus better expressing users' true preferences. In this paper, we aim to leverage the questionnaire feedback to align the recommendation model with users' true preferences. However, due to the platform distribution mechanism and divergent user attitudes toward questionnaires, the questionnaire feedback data frequently becomes sparse and exhibits selection biases, resulting in challenges in feature integration and training process. To address these issues, we introduce a novel user Satisfaction Alignment framework that effectively leverages Questionnaire feedback to enhance Recommendation, named SAQRec. SAQRec begins by training an unbiased satisfaction model to impute satisfaction, addressing selection bias and data sparsity. Then, SAQRec aligns features with users' true preferences by disentangling satisfaction and dissatisfaction from click history and categorizing clicked items into multiple satisfaction levels through the imputed satisfactions. Additionally, the imputed satisfactions from the pre-trained unbiased satisfaction model serve as pseudo-labels to align the model's outputs with users' true preferences. Extensive experiments on both public and commercial datasets demonstrate SAQRec's superior integration of questionnaire feedback in recommendation models. Online A/B testing on a short video platform confirms its effectiveness in boosting user watch time and positive-to-negative feedback ratio, enhancing overall performance and user satisfaction.

In this paper, we address the issue of using logic rules to explain the results from legal case retrieval. The task is critical to legal case retrieval because the users (e.g., lawyers or judges) are highly specialized and require the system to provide logic, faithful, and interpretable explanations before making legal decisions. Recently, research efforts have been made to learn explainable legal case retrieval models. However, these methods usually select rationales (key sentences) from the legal cases as explanations, failing to provide faithful and logicly correct explanations. In this paper, we propose Neural-Symbolic enhanced Legal Case Retrieval (NS-LCR), a framework that explicitly conducts reasoning on the matching of legal cases through learning case-level and law-level logic rules. The learned rules are then integrated into the retrieval process in a neuro-symbolic manner. Benefiting from the logic and interpretable nature of the logic rules, NS-LCR is equipped with built-in faithful explainability. We also show that NS-LCR is a model-agnostic framework that can be plug-in for multiple legal retrieval models. To demonstrate the superiority of NS-LCR, we extend the benchmarks of LeCaRD and ELAM with manually annotated logic rules and propose a new explainability measure based on Large Language Models (LLMs). Extensive experiments show that NS-LCR can achieve state-of-the-art ranking performances, and the empirical analysis also showed that NS-LCR is capable of providing faithful explanations for legal case retrieval.

In recommender systems, the cold-start problem poses a significant challenge, especially as users transition from being new to more engaged. Existing solutions often lack the granularity to accommodate this evolving user engagement, resulting in suboptimal performance for intermediate and older users. We introduce "Smooth Start", a unified model that addresses this oversight through a gating mechanism. This mechanism dynamically adjusts the information flow based on each user’s level of engagement, providing a tailored focus for more personalized recommendations. Experimental results validate the effectiveness of "Smooth Start" across varying levels of engagement, offering a more nuanced and efficient solution to the cold-start problem.

With the popularity of online music services, personalized music recommendation has garnered much research interest. Recommendation models are typically trained on datasets constructed from user feedback, which includes both the active feedback (e.g., clicking the Like or Skip buttons) and passive feedback (e.g., auto-play), with passive feedback comprising the majority. Due to the unavailability of user attention, the massive amount of passive feedback is unreliable, significantly compromising the quality of the training data. How to estimate the user's attention on the target music has become a critical problem in music recommendation. Heuristic methods such as exponential decay and negative sampling have been proposed. However, they either neglect the sequential dependencies between feedback actions or utilize only a small fraction of passive samples, leading to inaccurate and biased attention estimation. In this paper, we naturally propose modeling user attention prediction as a positive-unlabeled (PU) learning problem, where active feedback is treated as positive samples and passive feedback is treated as unlabeled samples, as we can only ensure that the user's attention is focused when she provides active feedback. Then we propose an extended PU-learning model with sequential dependencies, called UAE, which contains an unbiased user attention estimator and an unbiased propensity estimator. Subsequently, a joint learning algorithm is developed in which the attention and propensity estimators are optimized in alternating fashion. Theoretical analysis shows the unbiasedness and variance of the attention estimator and the propensity estimator. Extensive experiments on two large-scale datasets demonstrate the proposed UAE's effectiveness and generality in enhancing downstream music recommendation. One week online A/B testing on Huawei Music App manifests that UAE can significantly increase the users' play count and time over 2%, further demonstrating the effectiveness of UAE in real-world music recommendation products.

With the rapid advancements of large language models (LLMs), information retrieval (IR) systems, such as search engines and recommender systems, have undergone a significant paradigm shift. This evolution, while heralding new opportunities, introduces emerging challenges, particularly in terms of biases and unfairness, which may threaten the information ecosystem. In this paper, we present a comprehensive survey of existing works on emerging and pressing bias and unfairness issues in IR systems when the integration of LLMs. We first unify bias and unfairness issues as distribution mismatch problems, providing a groundwork for categorizing various mitigation strategies through distribution alignment. Subsequently, we systematically delve into the specific bias and unfairness issues arising from three critical stages of LLMs integration into IR systems: data collection, model development, and result evaluation. In doing so, we meticulously review and analyze recent literature, focusing on the definitions, characteristics, and corresponding mitigation strategies associated with these issues. Finally, we identify and highlight some open problems and challenges for future work, aiming to inspire researchers and stakeholders in the IR field and beyond to better understand and mitigate bias and unfairness issues of IR in this LLM era. We also consistently maintain a GitHub repository for the relevant papers and resources in this rising direction at https://github.com/KID-22/LLM-IR-Bias-Fairness-Survey.

Recently, the emergence of large language models (LLMs) has revolutionized the paradigm of information retrieval (IR) applications, especially in web search, by generating vast amounts of human-like texts on the Internet. As a result, IR systems in the LLM era are facing a new challenge: the indexed documents are now not only written by human beings but also automatically generated by the LLMs. How these LLM-generated documents influence the IR systems is a pressing and still unexplored question. In this work, we conduct a quantitative evaluation of IR models in scenarios where both human-written and LLM-generated texts are involved. Surprisingly, our findings indicate that neural retrieval models tend to rank LLM-generated documents higher. We refer to this category of biases in neural retrievers towards the LLM-generated content as the source bias. Moreover, we discover that this bias is not confined to the first-stage neural retrievers, but extends to the second-stage neural re-rankers. Then, in-depth analyses from the perspective of text compression indicate that LLM-generated texts exhibit more focused semantics with less noise, making it easier for neural retrieval models to semantic match. To mitigate the source bias, we also propose a plug-and-play debiased constraint for the optimization objective, and experimental results show its effectiveness. Finally, we discuss the potential severe concerns stemming from the observed source bias and hope our findings can serve as a critical wake-up call to the IR community and beyond. To facilitate future explorations of IR in the LLM era, the constructed two new benchmarks are available at https://github.com/KID-22/Source-Bias.

In recent years, graph contrastive learning (GCL) has received increasing attention in recommender systems due to its effectiveness in reducing bias caused by data sparsity. However, most existing GCL models rely on heuristic approaches and usually assume entity independence when constructing contrastive views. We argue that these methods struggle to strike a balance between semantic invariance and view hardness across the dynamic training process, both of which are critical factors in graph contrastive learning. To address the above issues, we propose a novel GCL-based recommendation framework RGCL, which effectively maintains the semantic invariance of contrastive pairs and dynamically adapts as the model capability evolves through the training process. Specifically, RGCL first introduces decision boundary-aware adversarial perturbations to constrain the exploration space of contrastive augmented views, avoiding the decrease of task-specific information. Furthermore, to incorporate global user-user and item-item collaboration relationships for guiding on the generation of hard contrastive views, we propose an adversarial-contrastive learning objective to construct a relation-aware view-generator. Besides, considering that unsupervised GCL could potentially narrower margins between data points and the decision boundary, resulting in decreased model robustness, we introduce the adversarial examples based on maximum perturbations to achieve margin maximization. We also provide theoretical analyses on the effectiveness of our designs. Through extensive experiments on five public datasets, we demonstrate the superiority of RGCL compared against twelve baseline models.

Reciprocal recommender systems~(RRS), conducting bilateral recommendations between two involved parties, have gained increasing attention for enhancing matching efficiency. However, the majority of existing methods in the literature still reuse conventional ranking metrics to separately assess the performance on each side of the recommendation process. These methods overlook the fact that the ranking outcomes of both sides collectively influence the effectiveness of the RRS, neglecting the necessity of a more holistic evaluation and a capable systemic solution.

In video recommendation, an ongoing effort is to satisfy users' personalized information needs by leveraging their logged watch time. However, watch time prediction suffers from duration bias, hindering its ability to reflect users' interests accurately. Existing label-correction approaches attempt to uncover user interests through grouping and normalizing observed watch time according to video duration. Although effective to some extent, we found that these approaches regard completely played records (i.e., a user watches the entire video) as equally high interest, which deviates from what we observed on real datasets: users have varied explicit feedback proportion when completely playing videos. In this paper, we introduce the counterfactual watch time (CWT), the potential watch time a user would spend on the video if its duration is sufficiently long. Analysis shows that the duration bias is caused by the truncation of CWT due to the video duration limitation, which usually occurs on those completely played records. Besides, a Counterfactual Watch Model (CWM) is proposed, revealing that CWT equals the time users get the maximum benefit from video recommender systems. Moreover, a cost-based transform function is defined to transform the CWT into the estimation of user interest, and the model can be learned by optimizing a counterfactual likelihood function defined over observed user watch times. Extensive experiments on three real video recommendation datasets and online A/B testing demonstrated that CWM effectively enhanced video recommendation accuracy and counteracted the duration bias.

Recommender systems have been widely used in e-commerce, and re-ranking models are playing an increasingly significant role in the domain, which leverages the inter-item influence and determines the final recommendation lists. Online learning methods keep updating a deployed model with the latest available samples to capture the shifting of the underlying data distribution in e-commerce. However, they depend on the availability of real user feedback, which may be delayed by hours or even days, such as item purchases, leading to a lag in model enhancement. In this paper, we propose a novel extension of online learning methods for re-ranking modeling, which we term LAST, an acronym for Learning At Serving Time. It circumvents the requirement of user feedback by using a surrogate model to provide the instructional signal needed to steer model improvement. Upon receiving an online request, LAST finds and applies a model modification on the fly before generating a recommendation result for the request. The modification is request-specific and transient. It means the modification is tailored to and only to the current request to capture the specific context of the request. After a request, the modification is discarded, which helps to prevent error propagation and stabilizes the online learning procedure since the predictions of the surrogate model may be inaccurate. Most importantly, as a complement to feedback-based online learning methods, LAST can be seamlessly integrated into existing online learning systems to create a more adaptive and responsive recommendation experience. Comprehensive experiments, both offline and online, affirm that LAST outperforms state-of-the-art re-ranking models.

Fair re-ranking aims to redistribute ranking slots among items more equitably to ensure responsibility and ethics. The exploration of redistribution problems has a long history in economics, offering valuable insights for conceptualizing fair re-ranking as a taxation process. Such a formulation provides us with a fresh perspective to re-examine fair re-ranking and inspire the development of new methods. From a taxation perspective, we theoretically demonstrate that most previous fair re-ranking methods can be reformulated as an item-level tax policy. Ideally, a good tax policy should be effective and conveniently controllable to adjust ranking resources. However, both empirical and theoretical analyses indicate that the previous item-level tax policy cannot meet two ideal controllable requirements: (1) continuity, ensuring minor changes in tax rates result in small accuracy and fairness shifts; (2) controllability over accuracy loss, ensuring precise estimation of the accuracy loss under a specific tax rate. To overcome these challenges, we introduce a new fair re-ranking method named Tax-rank, which levies taxes based on the difference in utility between two items. Then, we efficiently optimize such an objective by utilizing the Sinkhorn algorithm in optimal transport. Upon a comprehensive analysis, Our model Tax-rank offers a superior tax policy for fair re-ranking, theoretically demonstrating both continuity and controllability over accuracy loss. Experimental results show that Tax-rank outperforms all state-of-the-art baselines on two ranking tasks.

In real-world recommender systems, such as in the music domain, repeat consumption is a common phenomenon where users frequently listen to a small set of preferred songs or artists repeatedly. The key point of modeling repeat consumption is capturing the temporal patterns between a user's repeated consumption of the items. Existing studies often rely on heuristic assumptions, such as assuming an exponential distribution for the temporal gaps. However, due to the high complexity of real-world recommender systems, these pre-defined distributions may fail to capture the intricate dynamic user consumption patterns, leading to sub-optimal performance. Drawing inspiration from the flexibility of neural ordinary differential equations (ODE) in capturing the dynamics of complex systems, we propose ReCODE, a novel model-agnostic framework that utilizes neural ODE to model repeat consumption. ReCODE comprises two essential components: a user's static preference prediction module and the modeling of user dynamic repeat intention. By considering both immediate choices and historical consumption patterns, ReCODE offers comprehensive modeling of user preferences in the target context. Moreover, ReCODE seamlessly integrates with various existing recommendation models, including collaborative-based and sequential-based models, making it easily applicable in different scenarios. Experimental results on two real-world datasets consistently demonstrate that ReCODE significantly improves the performance of base models and outperforms other baseline methods.

Legal document retrieval and judgment prediction are crucial tasks in intelligent legal systems. In practice, determining whether two documents share the same judgments is essential for establishing their relevance in legal retrieval. However, existing legal retrieval studies either ignore the vital role of judgment prediction or rely on implicit training objectives, expecting a proper alignment of legal documents in vector space based on their judgments. Neither approach provides explicit evidence of judgment consistency for relevance modeling, leading to inaccuracies and a lack of transparency in retrieval. To address this issue, we propose a law-guided method, namely GEAR, within the generative retrieval framework. GEAR explicitly integrates judgment prediction with legal document retrieval in a sequence-to-sequence manner. Specifically, given the intricate nature of legal documents, we first extract rationales from documents based on the definition of charges in law. We then employ these rationales as queries, ensuring efficiency and producing a shared, informative document representation for both tasks. Second, in accordance with the inherent hierarchy of law, we construct a law structure constraint tree and represent each candidate document as a hierarchical semantic ID based on this tree. This empowers GEAR to perform dual predictions for judgment and relevant documents in a single inference, i.e., traversing the tree from the root through intermediate judgment nodes, to document-specific leaf nodes. Third, we devise the revision loss that jointly minimizes the discrepancy between the IDs of predicted and labeled judgments, as well as retrieved documents, thus improving accuracy and consistency for both tasks. Extensive experiments on two Chinese legal case retrieval datasets show the superiority of GEAR over state-of-the-art methods while maintaining competitive judgment prediction performance. Moreover, we validate the effectiveness of GEAR on a French statutory article retrieval dataset, reaffirming its robustness across languages and domains.

Nowadays, many platforms provide users with both search and recommendation services as important tools for accessing information. The phenomenon has led to a correlation between user search and recommendation behaviors, providing an opportunity to model user interests in a fine-grained way. Existing approaches either model user search and recommendation behaviors separately or overlook the different transitions between user search and recommendation behaviors. In this paper, we propose a framework named UniSAR that effectively models the different types of fine-grained behavior transitions for providing users a Unified Search And Recommendation service. Specifically, UniSAR models the user transition behaviors between search and recommendation through three steps: extraction, alignment, and fusion, which are respectively implemented by transformers equipped with pre-defined masks, contrastive learning that aligns the extracted fine-grained user transitions, and cross-attentions that fuse different transitions. To provide users with a unified service, the learned representations are fed into the downstream search and recommendation models. Joint learning on both search and recommendation data is employed to utilize the knowledge and enhance each other. Experimental results on two public datasets demonstrated the effectiveness of UniSAR in terms of enhancing both search and recommendation simultaneously. The experimental analysis further validates that UniSAR enhances the results by successfully modeling the user transition behaviors between search and recommendation.

Incorporating Search and Recommendation (S&R) services within a singular application is prevalent in online platforms, leading to a new task termed open-app motivation prediction, which aims to predict whether users initiate the application with the specific intent of information searching, or to explore recommended content for entertainment. Studies have shown that predicting users' motivation to open an app can help to improve user engagement and enhance performance in various downstream tasks. However, accurately predicting open-app motivation is not trivial, as it is influenced by user-specific factors, search queries, clicked items, as well as their temporal occurrences. Furthermore, these activities occur sequentially and exhibit intricate temporal dependencies. Inspired by the success of the Neural Hawkes Process (NHP) in modeling temporal dependencies in sequences, this paper proposes a novel neural Hawkes process model to capture the temporal dependencies between historical user browsing and querying actions. The model, referred to as Neural Hawkes Process-based Open-App Motivation prediction model (NHP-OAM), employs a hierarchical transformer and a novel intensity function to encode multiple factors, and open-app motivation prediction layer to integrate time and user-specific information for predicting users' open-app motivations. To demonstrate the superiority of our NHP-OAM model and construct a benchmark for the Open-App Motivation Prediction task, we not only extend the public S&R dataset ZhihuRec but also construct a new real-world Open-App Motivation Dataset (OAMD). Experiments on these two datasets validate NHP-OAM's superiority over baseline models. Further downstream application experiments demonstrate NHP-OAM's effectiveness in predicting users' Open-App Motivation, highlighting the immense application value of NHP-OAM.

With the application of generation models, internet is increasingly inundated with AI-generated content (AIGC), causing both real and AI-generated content indexed in corpus for search. This paper explores the impact of AI-generated images on text-image search in this scenario. Firstly, we construct a benchmark consisting of both real and AI-generated images for this study. In this benchmark, AI-generated images possess visual semantics sufficiently similar to real images. Experiments on this benchmark reveal that text-image retrieval models tend to rank the AI-generated images higher than the real images, even though the AI-generated images do not exhibit more visually relevant semantics to the queries than real images. We call this bias as invisible relevance bias. This bias is detected across retrieval models with different training data and architectures. Further exploration reveals that mixing AI-generated images into the training data of retrieval models exacerbates the invisible relevance bias. These problems cause a vicious cycle in which AI-generated images have a higher chance of exposing from massive data, which makes them more likely to be mixed into the training of retrieval models and such training makes the invisible relevance bias more and more serious. To mitigate this bias and elucidate the potential causes of the bias, firstly, we propose an effective method to alleviate this bias. Subsequently, we apply our proposed debiasing method to retroactively identify the causes of this bias, revealing that the AI-generated images induce the image encoder to embed additional information into their representation. This information makes the retriever estimate a higher relevance score. We conduct experiments to support this assertion.

Reinforcement learning (RL) has gained popularity in recommender systems for improving long-term performance by effectively exploring users' interests. However, modern recommender systems face the challenge of different user behavioral patterns among millions of items, making exploration more difficult. For example, users with varying activity levels require different exploration intensities. Unfortunately, previous studies often overlook this aspect and apply a uniform exploration strategy to all users, which ultimately hampers long-term user experiences. To tackle these challenges, we propose User-Oriented Exploration Policy (UOEP), a novel approach that enables fine-grained exploration among user groups. We first construct a distributional critic that allows policy optimization based on varying quantile levels of cumulative reward feedback from users, representing user groups with different activity levels. Using this critic as a guide, we design a population of distinct actors dedicated to effective and fine-grained exploration within their respective user groups. To simultaneously enhance diversity and stability during the exploration process, we also introduce a population-level diversity regularization term and a supervision module. Experimental results on public recommendation datasets validate the effectiveness of our approach, as it outperforms all other baselines in terms of long-term performance. Moreover, further analyses reveal the benefits of our approach, including improved performance for low-activity users and increased fairness among users.

The results of information retrieval (IR) are usually presented in the form of a ranking list of candidate documents, such as web search for humans and retrieval-augmented generation for large language models (LLMs). List-aware retrieval aims to capture the list-level contextual features to return a better list, mainly including reranking and truncation. Reranking finely re-scores the documents in the list. Truncation dynamically determines the cut-off point of the ranked list to achieve the trade-off between overall relevance and avoiding misinformation from irrelevant documents. Previous studies treat them as two separate tasks and model them separately. However, the separation is not optimal. First, it is hard to share the contextual information of the ranking list between the two tasks. Second, the separate pipeline usually meets the error accumulation problem, where the small error from the reranking stage can largely affect the truncation stage. To solve these problems, we propose a Reranking-Truncation joint model (GenRT) that can perform the two tasks concurrently. GenRT integrates reranking and truncation via a generative paradigm based on an encoder-decoder architecture with novel loss functions for joint optimization to learn both tasks. Sharing parameters by the joint model is conducive to making full use of the common modeling information of the two tasks. Besides, the two tasks are performed concurrently and co-optimized to solve the error accumulation problem between separate stages. Experiments on public learning-to-rank benchmarks and open-domain Q&A tasks show that our method achieves SOTA performance on both reranking and truncation tasks for web search and retrieval-augmented LLMs.

In pursuit of fairness and balanced development, recommender systems (RS) often prioritize group fairness, ensuring that specific groups maintain a minimum level of exposure over a given period. For example, RS platforms aim to ensure adequate exposure for new providers or specific categories of items according to their needs. Modern industry RS usually adopts a two-stage pipeline: stage-1 (retrieval stage) retrieves hundreds of candidates from millions of items distributed across various servers, and stage-2 (ranking stage) focuses on presenting a small-size but accurate selection from items chosen in stage-1. Existing efforts for ensuring amortized group exposures focus on stage-2, however, stage-1 is also critical for the task. Without a high-quality set of candidates, the stage-2 ranker cannot ensure the required exposure of groups. Previous fairness-aware works designed for stage-2 typically require accessing and traversing all items. In stage-1, however, millions of items are distributively stored in servers, making it infeasible to traverse all of them. How to ensure group exposures in the distributed retrieval process is a challenging question. To address this issue, we introduce a model named FairSync, which transforms the problem into a constrained distributed optimization problem. Specifically, FairSync resolves the issue by moving it to the dual space, where a central node aggregates historical fairness data into a vector and distributes it to all servers. To trade off the efficiency and accuracy, the gradient descent technique is used to periodically update the parameter of the dual vector. The experiment results on two public recommender retrieval datasets showcased that FairSync outperformed all the baselines, achieving the desired minimum level of exposures while maintaining a high level of retrieval accuracy.

Providing human-understandable explanations for the matching predictions is still challenging for current legal case matching methods. One difficulty is that legal cases are semi-structured text documents with complicated case-case and case-law article correlations. To tackle the issue, we propose a novel graph optimal transport (GOT)-based legal case matching model that is able to provide not only the matching predictions but also plausible and faithful explanations for the prediction. The model, called GEIOT-Match, first constructs a heterogeneous graph to explicitly represent the semi-structured nature of legal cases and their associations with the law articles. Therefore, matching two legal cases amounts to identifying the rationales from the paired legal case sub-graphs in the heterogeneous graph and then aligning between them. An inverse optimal transport (IOT) model on graphs is learned to extract rationales from paired legal cases. The extracted rationales and the heterogeneous graph demonstrate the key legal characteristics of legal cases, which can be further used to conduct matching and generate explanations for the matching. Experimental results showed that GEIOT-Match outperformed state-of-the-art baselines in terms of matching prediction, rationale extraction, and natural language explanation generation.

Group-fair recommendation aims at ensuring the equality of recommendation results across user groups categorized by sensitive attributes (e.g., gender, occupation, etc.). Existing group-fair recommendation models traditionally employ original user embeddings for both training and testing, primarily focusing on statistical learning while imposing group fairness constraints under the I.I.D. assumption. However, these models encounter limitations when addressing out-of-distribution (OOD) sensitive attributes. The fundamental issue of unfairness within user embeddings arises from a causal perspective, where each embedding vector comprises an exogenous component devoid of correlations with sensitive attributes and an endogenous component strongly correlated with these attributes. Overlooking the distinction between these two components during model training renders models sensitive to shifts in the distribution of sensitive attributes. This paper introduces the concept of Counterfactual Group Fairness (CGF) along with a corresponding metric to evaluate group fairness in scenarios involving OOD sensitive attributes in recommender systems. Building on this foundation, we propose a model-agnostic causal embedding learning framework named MACE. MACE effectively disentangles user embedding vectors into their exogenous and endogenous parts, thus ensuring group fairness, even in the presence of OOD sensitive attributes in embeddings. Specifically, MACE identifies the exogenous part of each user's embedding using mutual information minimization, treating it as instrumental variables. Subsequently, under the constraint of CGF, MACE reconstructs the endogenous and exogenous parts using the instrumental variable regression, combines the obtained parts into novel user embeddings using deep neural networks, and uses the combined embeddings for fair recommendation. Experimental results demonstrated that MACE can outperform the state-of-the-art baselines in terms of the metric of CGF while maintaining a comparable recommendation accuracy.

Matching two natural language sentences is a fundamental problem in both natural language processing and information retrieval. Preliminary studies have shown that the syntactic structures help improve the matching accuracy, and different syntactic structures in natural language are complementary to sentence semantic understanding. Ideally, a matching model would leverage all syntactic information. Existing models, however, are only able to combine limited (usually one) types of syntactic information due to the complex and heterogeneous nature of the syntactic information. To deal with the problem, we propose a novel matching model, which formulates sentence matching as a representation learning task on a syntactic-informed heterogeneous graph. The model, referred to as SIGN (Syntactic-Informed Graph Network), first constructs a heterogeneous matching graph based on the multiple syntactic structures of two input sentences. Then the graph attention network algorithm is applied to the matching graph to learn the high-level representations of the nodes. With the help of the graph learning framework, the multiple syntactic structures, as well as the word semantics, can be represented and interacted in the matching graph and therefore collectively enhance the matching accuracy. We conducted comprehensive experiments on three public datasets. The results demonstrate that SIGN outperforms the state of the art and also can discriminate the sentences in an interpretable way.

In online industrial advertising systems, conversion actions (e.g., purchases or downloads) often occur significantly delayed, even up to several days or weeks after the user clicks. This phenomenon leads to the crucial challenge calleddelayed feedback problem in streaming CVR prediction, that is, the online systems cannot receive the true label of conversions immediately for continuous training. To mitigate the delayed feedback problem, recent state-of-the-art methods often apply sample duplicate mechanisms to introduce early certain conversion information. Nevertheless, these works have overlooked a crucial issue of rapid shifts in data distribution and considered both the newly observed data and duplicated early data together, resulting in biases in both distributions. In this work, we propose a Dually enhanced Delayed Feedback Model (DDFM), which tackles the above issues by treating the newly observed data and duplicated early data separately. DDFM consists of dual unbiased CVR estimators that share the same form but utilize different latent variables as weights: one for the newly observed data and the other for the duplicated early data. To avoid high variance, we adopt an addition-only formula for these latent variables, eliminating multiplication or division operations. Furthermore, we design a shared-bottom network that efficiently and jointly estimates the latent variables in DDFM. Theoretical analysis demonstrates the unbiasedness and convergence properties of DDFM. Extensive experiments on both public and industrial large-scale real-world datasets exhibit that our proposed DDFM consistently outperforms existing state-of-the-art methods.

In real-world streaming recommender systems, user preferences often dynamically change over time (e.g., a user may have different preferences during weekdays and weekends). Existing bandit-based streaming recommendation models only consider time as a timestamp, without explicitly modeling the relationship between time variables and time-varying user preferences. This leads to recommendation models that cannot quickly adapt to dynamic scenarios. To address this issue, we propose a contextual bandit approach using hypernetwork, called HyperBandit, which takes time features as input and dynamically adjusts the recommendation model for time-varying user preferences. Specifically, HyperBandit maintains a neural network capable of generating the parameters for estimating time-varying rewards, taking into account the correlation between time features and user preferences. Using the estimated time-varying rewards, a bandit policy is employed to make online recommendations by learning the latent item contexts. To meet the real-time requirements in streaming recommendation scenarios, we have verified the existence of a low-rank structure in the parameter matrix and utilize low-rank factorization for efficient training. Theoretically, we demonstrate a sublinear regret upper bound against the best policy. Extensive experiments on real-world datasets show that the proposed HyperBandit consistently outperforms the state-of-the-art baselines in terms of accumulated rewards.

The confluence of Search and Recommendation (S&R) services is vital to online services, including e-commerce and video platforms. The integration of S&R modeling is a highly intuitive approach adopted by industry practitioners. However, there is a noticeable lack of research conducted in this area within academia, primarily due to the absence of publicly available datasets. Consequently, a substantial gap has emerged between academia and industry regarding research endeavors in joint optimization using user behavior data from both S&R services. To bridge this gap, we introduce the first large-scale, real-world dataset KuaiSAR of integrated Search And Recommendation behaviors collected from Kuaishou, a leading short-video app in China with over 350 million daily active users. Previous research in this field has predominantly employed publicly available semi-synthetic datasets, with artificially fabricated search behaviors. Distinct from previous datasets, KuaiSAR contains genuine user behaviors, including the occurrence of each interaction within either search or recommendation service, and the users' transitions between the two services. This work aids in joint modeling of S&R, and utilizing search data for recommender systems (and recommendation data for search engines). Furthermore, due to the various feedback labels associated with user-video interactions, KuaiSAR also supports a broad range of tasks, including intent recommendation, multi-task learning, and modeling of long sequential multi-behavioral patterns. We believe this dataset will serve as a catalyst for innovative research and bridge the gap between academia and industry in understanding the S&R services in practical, real-world applications. The dataset is available at https://ethan00si.github.io/KuaiSAR/. The dataset is also shared at https://zenodo.org/record/8181109.

Multi-stage ranking pipelines have become widely used strategies in modern recommender systems, where the final stage aims to return a ranked list of items that balances a number of requirements such as user preference, diversity, novelty etc. Linear scalarization is arguably the most widely used technique to merge multiple requirements into one optimization objective, by summing up the requirements with certain preference weights. Existing final-stage ranking methods often adopt a static model where the preference weights are determined during offline training and kept unchanged during online serving. Whenever a modification of the preference weights is needed, the model has to be re-trained, which is time and resources inefficient. Meanwhile, the most appropriate weights may vary greatly for different groups of targeting users or at different time periods (e.g., during holiday promotions). In this paper, we propose a framework called controllable multi-objective re-ranking (CMR) which incorporates a hypernetwork to generate parameters for a re-ranking model according to different preference weights. In this way, CMR is enabled to adapt the preference weights according to the environment changes in an online manner, without retraining the models. Moreover, we classify practical business-oriented tasks into four main categories and seamlessly incorporate them in a new proposed re-ranking model based on an Actor-Evaluator framework, which serves as a reliable real-world testbed for CMR. Offline experiments based on the dataset collected from Taobao App showed that CMR improved several popular re-ranking models by using them as underlying models. Online A/B tests also demonstrated the effectiveness and trustworthiness of CMR.

Contextual batched bandit (CBB) is a setting where a batch of rewards is observed from the environment at the end of each episode, but the rewards of the non-executed actions are unobserved, resulting in partial-information feedback. Existing approaches for CBB often ignore the rewards of the non-executed actions, leading to underutilization of feedback information. In this paper, we propose an efficient approach called Sketched Policy Updating with Imputed Rewards (SPUIR) that completes the unobserved rewards using sketching, which approximates the full-information feedbacks. We formulate reward imputation as an imputation regularized ridge regression problem that captures the feedback mechanisms of both executed and non-executed actions. To reduce time complexity, we solve the regression problem using randomized sketching. We prove that our approach achieves an instantaneous regret with controllable bias and smaller variance than approaches without reward imputation. Furthermore, our approach enjoys a sublinear regret bound against the optimal policy. We also present two extensions, a rate-scheduled version and a version for nonlinear rewards, making our approach more practical. Experimental results show that SPUIR outperforms state-of-the-art baselines on synthetic, public benchmark, and real-world datasets.

The debut of ChatGPT has recently attracted significant attention from the natural language processing (NLP) community and beyond. Existing studies have demonstrated that ChatGPT shows significant improvement in a range of downstream NLP tasks, but the capabilities and limitations of ChatGPT in terms of recommendations remain unclear. In this study, we aim to enhance ChatGPT’s recommendation capabilities by aligning it with traditional information retrieval (IR) ranking capabilities, including point-wise, pair-wise, and list-wise ranking. To achieve this goal, we re-formulate the aforementioned three recommendation policies into prompt formats tailored specifically to the domain at hand. Through extensive experiments on four datasets from different domains, we analyze the distinctions among the three recommendation policies. Our findings indicate that ChatGPT achieves an optimal balance between cost and performance when equipped with list-wise ranking. This research sheds light on a promising direction for aligning ChatGPT with recommendation tasks. To facilitate further explorations in this area, the full code and detailed original results are open-sourced at https://github.com/rainym00d/LLM4RS.

In the video recommendation, watch time is commonly adopted as an indicator of user interest. However, watch time is not only influenced by the matching of users’ interests but also by other factors, such as duration bias and noisy watching. Duration bias refers to the tendency for users to spend more time on videos with longer durations, regardless of their actual interest level. Noisy watching, on the other hand, describes users taking time to determine whether they like a video or not, which can result in users spending time watching videos they do not like. Consequently, the existence of duration bias and noisy watching make watch time an inadequate label for indicating user interest. Furthermore, current methods primarily address duration bias and ignore the impact of noisy watching, which may limit their effectiveness in uncovering user interest from watch time. In this study, we first analyze the generation mechanism of users’ watch time from a unified causal viewpoint. Specifically, we considered the watch time as a mixture of the user’s actual interest level, the duration-biased watch time, and the noisy watch time. To mitigate both the duration bias and noisy watching, we propose Debiased and Denoised watch time Correction (D2Co), which can be divided into two steps: First, we employ a duration-wise Gaussian Mixture Model plus frequency-weighted moving average for estimating the bias and noise terms; then we utilize a sensitivity-controlled correction function to separate the user interest from the watch time, which is robust to the estimation error of bias and noise terms. The experiments on two public video recommendation datasets and online A/B testing indicate the effectiveness of the proposed method.

Modern online service providers such as online shopping platforms often provide both search and recommendation (S&R) services to meet different user needs. Rarely has there been any effective means of incorporating user behavior data from both S&R services. Most existing approaches either simply treat S&R behaviors separately, or jointly optimize them by aggregating data from both services, ignoring the fact that user intents in S&R can be distinctively different. In our paper, we propose a Search-Enhanced framework for the Sequential Recommendation (SESRec) that leverages users' search interests for recommendation, by disentangling similar and dissimilar representations within S&R behaviors. Specifically, SESRec first aligns query and item embeddings based on users' query-item interactions for the computations of their similarities. Two transformer encoders are used to learn the contextual representations of S&R behaviors independently. Then a contrastive learning task is designed to supervise the disentanglement of similar and dissimilar representations from behavior sequences of S&R. Finally, we extract user interests by the attention mechanism from three perspectives, i.e., the contextual representations, the two separated behaviors containing similar and dissimilar interests. Extensive experiments on both industrial and public datasets demonstrate that SESRec consistently outperforms state-of-the-art models. Empirical studies further validate that SESRec successfully disentangle similar and dissimilar user interests from their S&R behaviors.

Legal case matching, which automatically constructs a model to estimate the similarities between the source and target cases, has played an essential role in intelligent legal systems. Semantic text matching models have been applied to the task where the source and target legal cases are considered as long-form text documents. These general-purpose matching models make the predictions solely based on the texts in the legal cases, overlooking the essential role of the law articles in legal case matching. In the real world, the matching results (e.g., relevance labels) are dramatically affected by the law articles because the contents and the judgments of a legal case are radically formed on the basis of law. From the causal sense, a matching decision is affected by the mediation effect from the cited law articles by the legal cases, and the direct effect of the key circumstances (e.g., detailed fact descriptions) in the legal cases. In light of the observation, this paper proposes a model-agnostic causal learning framework called Law-Match, under which the legal case matching models are learned by respecting the corresponding law articles. Given a pair of legal cases and the related law articles, Law-Match considers the embeddings of the law articles as instrumental variables(IVs), and the embeddings of legal cases as treatments. Using IV regression, the treatments can be decomposed into law-related and law-unrelated parts, respectively reflecting the mediation and direct effects. These two parts are then combined with different weights to collectively support the final matching prediction. We show that the framework is model-agnostic, and a number of legal case matching models can be applied as the underlying models. Comprehensive experiments show that Law-Match can outperform state-of-the-art baselines on three public datasets.

Identifying the causes of an event, also termed as causal attribution, is a commonly encountered task in many application problems. Available methods, mostly in Bayesian or causal inference literature, suffer from two main drawbacks: 1) cannot attribute for individuals, and 2) attributing one single cause at a time and cannot deal with the interaction effect among multiple causes. In this paper, based on our proposed new measurement, called conditional counterfactual causal effect (CCCE), we introduce an individual causal attribution method, which is able to utilize the individual observation as the evidence and consider common influence and interaction effect of multiple causes simultaneously. We discuss the identifiability of CCCE and also give the identification formulas under proper assumptions. Finally, we conduct experiments on simulated and real data to illustrate the effectiveness of CCCE and the results show that our proposed method outperforms significantly state-of-the-art methods.

Alleviating the examination and trust bias in ranking systems is an important research line in unbiased learning-to-rank (ULTR). Current methods typically use the propensity to correct the biased user clicks and then learn ranking models based on the corrected clicks. Though successes have been achieved, directly modifying the clicks suffers from the inherent high variance because the propensities are usually involved in the denominators of corrected clicks. The problem gets even worse in the situation of mixed examination and trust bias. To address the issue, this paper proposes a novel ULTR method called Decomposed Ranking Debiasing (DRD). DRD is tailored for learning unbiased relevance models with low variance in the existence of examination and trust bias. Unlike existing methods that directly modify the original user clicks, DRD proposes to decompose each click prediction as the combination of a relevance term outputted by the ranking model and other bias terms. The unbiased relevance model, therefore, can be learned by fitting the overall click predictions to the biased user clicks. A joint learning algorithm is developed to learn the relevance and bias models' parameters alternatively. Theoretical analysis showed that, compared with existing methods, DRD has lower variance while retains unbiasedness. Empirical studies indicated that DRD can effectively reduce the variance and outperform the state-of-the-art ULTR baselines.

In this paper, we address the issue of recommending fairly from the aspect of providers, which has become increasingly essential in multistakeholder recommender systems. Existing studies on provider fairness usually focused on designing proportion fairness (PF) metrics that first consider systematic fairness. However, sociological researches show that to make the market more stable, max-min fairness (MMF) is a better metric. The main reason is that MMF aims to improve the utility of the worst ones preferentially, guiding the system to support the providers in weak market positions. When applying MMF to recommender systems, how to balance user preferences and provider fairness in an online recommendation scenario is still a challenging problem. In this paper, we proposed an online re-ranking model named Provider Max-min Fairness Re-ranking (P-MMF) to tackle the problem. Specifically, P-MMF formulates provider fair recommendation as a resource allocation problem, where the exposure slots are considered the resources to be allocated to providers and the max-min fairness is used as the regularizer during the process. We show that the problem can be further represented as a regularized online optimizing problem and solved efficiently in its dual space. During the online re-ranking phase, a momentum gradient descent method is designed to conduct the dynamic re-ranking. Theoretical analysis showed that the regret of P-MMF can be bounded. Experimental results on four public recommender datasets demonstrated that P-MMF can outperformed the state-of-the-art baselines. Experimental results also show that P-MMF can retain small computationally costs on a corpus with the large number of items.

Recommender systems are currently widely used in various applications helping people filter information. Existing models always embed the rich information for recommendation, such as items, users, and contexts in real-value vectors, and make predictions based on these vectors. In the view of causal inference, the associations between representation vectors and user feedback are inevitably a mixture of the causal part that describes why a user prefers an item, and the non-causal part that merely reflects the statistical dependencies, for example, the display ranking position and sales promotion. However, most recommender systems assume the user-item interactions are only affected by user preferences, neglecting the striking differences between these two associations. To address this problem, we propose a model-agnostic causal learning framework called IV4Rec+ that can effectively decompose the embedding vectors into these two parts. Moreover, two strategies are proposed to utilize search queries as instrumental variables: IV4Rec+(I) only decomposes the item embeddings, while IV4Rec+(UI) decomposes both user and item embeddings. IV4Rec+ is a model-agnostic design that can be applied to many existing recommender systems, e.g., DIN, NRHUB, and SRGNN. Extensive experiments on three datasets show that IV4Rec+ significantly facilitates the performance of recommender systems and outperforms state-of-the-art frameworks.

Sequential recommender systems train their models based on a large amount of implicit user feedback data and may be subject to biases when users are systematically under/over-exposed to certain items. Unbiased learning based on inverse propensity scores (IPS), which estimate the probability of observing a user-item pair given the historical information, has been proposed to address the issue. In these methods, propensity score estimation is usually limited to the view of item, that is, treating the feedback data as sequences of items that interacted with the users. However, the feedback data can also be treated from the view of user, as the sequences of users that interact with the items. Moreover, the two views can jointly enhance the propensity score estimation. Inspired by the observation, we propose to estimate the propensity scores from the views of user and item, called Dually Enhanced Propensity Score Estimation (DEPS). Specifically, given a target user-item pair and the corresponding item and user interaction sequences, DEPS first constructs a time-aware causal graph to represent the user-item observational probability. According to the graph, two complementary propensity scores are estimated from the views of item and user, respectively, based on the same set of user feedback data. Finally, two transformers are designed to make use of the two propensity scores and make the final preference prediction. Theoretical analysis showed the unbiasedness and variance of DEPS. Experimental results on three publicly available benchmarks and a proprietary industrial dataset demonstrated that DEPS can significantly outperform the state-of-the-art baselines.

Studies have shown that the sentence’s syntactic structures are important for semantic sentence matching. A typical approach is encoding each sentence’s syntactic structure into an embedding vector, which can be combined with other features to predict the final matching scores. Though successes have been observed, embedding the whole syntactic structures as one vector inevitably overlooks the fine-grained syntax matching patterns, e.g. the alignment of specific term dependencies relations in the two inputted sentences. In this paper, we formalize the task of semantic sentence matching as a problem of graph matching in which each sentence is represented as a directed graph according to its syntactic structures. The syntax matching patterns (i.e. similar syntactic structures) between two sentences, therefore, can be extracted as the sub-graph structure alignments. The proposed method, referred to as Interacted Syntax Graphs (ISG), represents two sentences’ syntactic alignments as well as their semantic matching signals into one association graph. After that, the neural quadratic assignment programming (QAP) is adapted to extract syntactic matching patterns from the association graph. In this way, the syntactic structures fully interact in a fine granularity during the matching process. Experimental results on three public datasets demonstrated that ISG can outperform the state-of-the-art baselines effectively and efficiently. The empirical analysis also showed that ISG can match sentences in an interpretable way.

One typical approach to long-form document matching is first conducting alignment between cross-document sentence pairs, and then aggregating all of the sentence-level matching signals. However, this approach could be problematic because the alignment between documents is partial — despite two documents as a whole are well-matched, most of the sentences could still be dissimilar. Those dissimilar sentences lead to spurious sentence-level matching signals which may overwhelm the real ones, increasing the difficulties of learning the matching function. Therefore, accurately selecting the key sentences for document matching is becoming a challenging issue. To address the issue, we propose a novel matching approach that equips existing document matching models with an Optimal Partial Transport (OPT) based component, namely OPT-Match, which selects the sentences that play a major role in matching. Enjoying the partial transport properties of OPT, the selected key sentences can not only effectively enhance the matching accuracy, but also be explained as the rationales for the matching results. Extensive experiments on four publicly available datasets demonstrated that existing methods equipped with OPT-Match consistently outperformed the corresponding underlying methods. Evaluations also showed that the key sentences selected by OPT-Match were consistent with human-provided rationales.

In streaming media applications, like music Apps, songs are recommended in a continuous way in users' daily life. The recommended songs are played automatically although users may not pay any attention to them, posing a challenge of user attention bias in training recommendation models, i.e., the training instances contain a large number of false-positive labels (users' feedback). Existing approaches either directly use the auto-feedbacks or heuristically delete the potential false-positive labels. Both of the approaches lead to biased results because the false-positive labels cause the shift of training data distribution, hurting the accuracy of the recommendation models. In this paper, we propose a learning-based counterfactual approach to adjusting the user auto-feedbacks and learning the recommendation models using Neural Dueling Bandit algorithm, called NDB. Specifically, NDB maintains two neural networks: a user attention network for computing the importance weights that are used for modifying the original rewards, and another random network trained with dueling bandit for conducting online recommendations based on the modified rewards. Theoretical analysis showed that the modified rewards are statistically unbiased, and the learned bandit policy enjoys a sub-linear regret bound. Experimental results demonstrated that NDB can significantly outperform the state-of-the-art baselines.

As an essential operation of legal retrieval, legal case matching plays a central role in intelligent legal systems. This task has a high demand on the explainability of matching results because of its critical impacts on downstream applications --- the matched legal cases may provide supportive evidence for the judgments of target cases and thus influence the fairness and justice of legal decisions. Focusing on this challenging task, we propose a novel and explainable method, namely IOT-Match, with the help of computational optimal transport, which formulates the legal case matching problem as an inverse optimal transport (IOT) problem. Different from most existing methods, which merely focus on the sentence-level semantic similarity between legal cases, our IOT-Match learns to extract rationales from paired legal cases based on both semantics and legal characteristics of their sentences. The extracted rationales are further applied to generate faithful explanations and conduct matching. Moreover, the proposed IOT-Match is robust to the alignment label insufficiency issue commonly in practical legal case matching tasks, which is suitable for both supervised and semi-supervised learning paradigms. To demonstrate the superiority of our IOT-Match method and construct a benchmark of explainable legal case matching task, we not only extend the well-known Challenge of AI in Law (CAIL) dataset but also build a new Explainable Legal cAse Matching (ELAM) dataset, which contains lots of legal cases with detailed and explainable annotations. Experiments on these two datasets show that our IOT-Match outperforms state-of-the-art methods consistently on matching prediction, rationale extraction, and explanation generation.

Machine-learning based recommender system(RS) has become an effective means to help people automatically discover their interests. Existing models often represent the rich information for recommendation, such as items, users, and contexts, as embedding vectors and leverage them to predict users’ feedback. In the view of causal analysis, the associations between these embedding vectors and users’ feedback are a mixture of the causal part that describes why an item is preferred by a user, and the non-causal part that merely reflects the statistical dependencies between users and items, for example, the exposure mechanism, public opinions, display position, etc. However, existing RSs mostly ignored the striking differences between the causal parts and non-causal parts when using these embedding vectors. In this paper, we propose a model-agnostic framework named IV4Rec that can effectively decompose the embedding vectors into these two parts, hence enhancing recommendation results. Specifically, we jointly consider users’ behaviors in search scenarios and recommendation scenarios. Adopting the concepts in causal analysis, we embed users’ search behaviors as instrumental variables (IVs), to help decompose original embedding vectors in recommendation, i.e., treatments. IV4Rec then combines the two parts through deep neural networks and uses the combined results for recommendation. IV4Rec is model-agnostic and can be applied to a number of existing RSs such as DIN and NRHUB. Experimental results on both public and proprietary industrial datasets demonstrate that IV4Rec consistently enhances RSs and outperforms a framework that jointly considers search and recommendation.

Projecting the input text pair into a common semantic space where the matching function can be readily learned is an essential step for asymmetrical text matching. In the practice, it is often observed that the feature vectors from asymmetrical texts show a tendency to be gradually undistinguishable in the semantic space as the model is trained. However, the phenomenon is overlooked in existing studies. As a result, the feature vectors are constructed without any regularization, which inevitably hinders the learning of the downstream matching functions. In this paper, we first exploit the phenomenon and propose DDR-Match, a novel matching framework tailored for asymmetrical text matching. Specifically, in DDR-Match, a distribution distance-based regularizer is devised to accelerate the fusion of sequence representations corresponding to different domains in the semantic space. Then, we provide three instances of DDR-Match and make a comparison among them. DDR-Match is compatible with existing text matching methods by incorporating them as the underlying matching model. Four popular text matching methods are exploited in the paper. Extensive experimental results based on five publicly available benchmarks showed that DDR-Match consistently outperformed its underlying methods.

Regret bounds of online kernel selection in a finite kernel set have been well studied, having at least an order O( √ NT) of magnitude after T rounds, where N is the number of candidate kernels. But it is still an unsolved problem to achieve sublinear regret bounds of online kernel selection in a continuous kernel space under different learning frameworks. In this paper, to represent different learning frameworks of online kernel selection, we divide online kernel selection approaches in a continuous kernel space into two categories according to the order of selection and training at each round. Then we construct a surrogate hypothesis space that contains all the candidate kernels with bounded norms and inner products, representing the continuously varying hypothesis space. Finally, we decompose the regrets of the proposed online kernel selection categories into different types of instantaneous regrets in the surrogate hypothesis space, and derive optimal regret bounds of order O( √ T) of magnitude under mild assumptions, independent of the cardinality of the continuous kernel space. Empirical studies verified the correctness of the theoretical regret analyses.

Jingxuan Yang, Kerui Xu, Jun Xu, Si Li, Sheng Gao, Jun Guo, Nianwen Xue, Ji-Rong Wen. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Probability Ranking Principle (PRP)[31], which assumes that each document has a unique and independent probability to satisfy a particular information need, is one of the fundamental principles for ranking. Traditionally, heuristic ranking features and well-known learning-to-rank approaches have been designed by following the PRP principle. Recently, neural IR models, which adopt deep learning to enhance the ranking performances, also obey the PRP principle. Though it has been widely used for nearly five decades, in-depth analysis shows that PRP is not an optimal principle for ranking, due to its independent assumption that each document should be independent of the rest candidates. Counter examples include pseudo relevance feedback[24], interactive information retrieval[46], search result diversification[10] etc. To solve the problem, researchers recently proposed to model the dependencies among the documents during the designing of ranking models. A number of ranking models have been proposed and state-of-the-art ranking performances have been achieved. This tutorial aims to give a comprehensive survey on these recently developed ranking models that go beyond the PRP principle. The tutorial tries to categorize these models based on their intrinsic assumptions: assuming that the documents are independent, sequentially dependent, or globally dependent. In this way, we expect the researchers focusing on ranking in search and recommendation can have a novel angle of view on the designing of ranking models, and therefore can stimulate new ideas on developing novel ranking models.

The user feedbacks could be delayed in many streaming recommendation scenarios. As an example, the user feedbacks to a recommended coupon consist of the immediate feedback on the click event and the delayed feedback on the resultant conversion. The delayed feedbacks pose a challenge of training recommendation models using instances with incomplete labels. When being applied to real products, the challenge becomes more severe as the streaming recommendation models need to be retrained very frequently and the training instances need to be collected over very short time scales. Existing approaches either simply ignore the unobserved feedbacks or heuristically adjust the feedbacks on a static instance set, resulting in biases in the training data and hurting the accuracy of the learned recommenders. In this paper, we propose a novel and theoretic sound counterfactual approach to adjusting the user feedbacks and learning the recommendation models, called CBDF (Counterfactual Bandit with Delayed Feedback). CBDF formulates the streaming recommendation with delayed feedback as a problem of sequential decision making and models it with a batched bandit. To deal with the issue of delayed feedback, at each iteration (episode), a counterfactual importance sampling model is employed to re-weight the original feedbacks and generate the modified rewards. Based on the modified rewards, a batched bandit is learned for conducting online recommendation at the next iteration. Theoretical analysis showed that the modified rewards are statistically unbiased, and the learned bandit policy enjoys a sub-linear regret bound. Experimental results demonstrated that CBDF can outperform the state-of-the-art baselines on a synthetic dataset, the Criteo dataset, and a dataset from Tencent's WeChat app.

Probability Ranking Principle (PRP), which assumes that each document has a unique and independent probability to satisfy a particular information need, is one of the fundamental principles for ranking. Traditionally, heuristic ranking features and well-known learning-to-rank approaches have been designed by following the PRP principle. Recently, neural IR models, which adopt deep learning to enhance the ranking performances, also obey the PRP principle. Though it has been widely used for nearly five decades, in-depth analysis shows that PRP is not an optimal principle for ranking, due to its independent assumption that each document should be independent of the rest candidates. Counter examples include pseudo relevance feedback, interactive information retrieval, search result diversification, etc. To solve the problem, researchers recently proposed to model the dependencies among the documents during the designing of ranking models. A number of ranking models have been proposed and state-of-the-art ranking performances have been achieved. This tutorial aims to give a comprehensive survey on these recently developed ranking models that go beyond the PRP principle. The tutorial tries to categorize these models based on their intrinsic assumptions: assuming that the documents are independent, sequentially dependent, or globally dependent. In this way, we expect the researchers focusing on ranking in search and recommendation can have a novel angle of view on the designing of ranking models, and therefore can stimulate new ideas on developing novel ranking models. The material of this tutorial can be found in https://github.com/pl8787/wsdm2021-beyond-prp-tutorial.

This paper concerns user preference estimation in multi-round conversational recommender systems (CRS), which interacts with users by asking questions about attributes and recommending items multiple times in one conversation. Multi-round CRS such as EAR have been proposed in which the user's online feedback at both attribute level and item level can be utilized to estimate user preference and make recommendations. Though preliminary success has been shown, existing user preference models in CRS usually use the online feedback information as independent features or training instances, overlooking the relation between attribute-level and item-level feedback signals. The relation can be used to more precisely identify the reasons (e.g., some certain attributes) that trigger the rejection of an item, leading to more fine-grained utilization of the feedback information. To address aforementioned issue, this paper proposes a novel preference estimation model tailored for multi-round CRS, called Feedback-guided Preference Adaptation Network (FPAN). In FPAN, two gating modules are designed to respectively adapt the original user embedding and item-level feedback, both according to the online attribute-level feedback. The gating modules utilize the fine-grained attribute-level feedback to revise the user embedding and coarse-grained item-level feedback, achieving more accurate user preference estimation by considering the relation between feedback. Experimental results on two benchmarks showed that FPAN outperformed the state-of-the-art user preference models in CRS, and the multi-round CRS can also be enhanced by using FPAN as its recommender component.

Personalized search is a promising way to improve search qualities by taking user interests into consideration. Recently, machine learning and deep learning techniques have been successfully applied to search result personalization. Most existing models simply regard the personal search history as a static set of user behaviors and learn fixed ranking strategies based on all the recorded data. Though improvements have been achieved, the essence that the search process is a sequence of interactions between the search engine and user is ignored. The user’s interests may dynamically change during the search process, therefore, it would be more helpful if a personalized search model could track the whole interaction process and adjust its ranking strategy continuously. In this article, we adapt reinforcement learning to personalized search and propose a framework, referred to as RLPS. It utilizes a Markov Decision Process ( MDP ) to track sequential interactions between the user and search engine, and continuously update the underlying personalized ranking model with the user’s real-time feedback to learn the user’s dynamic interests. Within this framework, we implement two models: the listwise RLPS-L and the hierarchical RLPS-H. RLPS-L interacts with users and trains the ranking model with document lists, while RLPS-H improves model training by designing a layered structure and introducing document pairs. In addition, we also design a feedback-aware personalized ranking component to capture the user’s feedback, which impacts the user interest profile for the next query. Significant improvements over existing personalized search models are observed in the experiments on the public AOL search log and a commercial log.

One approach to matching texts from asymmetrical domains is projecting the input sequences into a common semantic space as feature vectors upon which the matching function can be readily defined and learned.In realworld matching practices, it is often observed that with the training goes on, the feature vectors projected from different domains tend to be indistinguishable.The phenomenon, however, is often overlooked in existing matching models.As a result, the feature vectors are constructed without any regularization, which inevitably increases the difficulty of learning the downstream matching functions.In this paper, we propose a novel match method tailored for text matching in asymmetrical domains, called WD-Match.In WD-Match, a Wasserstein distance-based regularizer is defined to regularize the features vectors projected from different domains.As a result, the method enforces the feature projection function to generate vectors such that those correspond to different domains cannot be easily discriminated.The training process of WD-Match amounts to a game that minimizes the matching loss regularized by the Wasserstein distance.WD-Match can be used to improve different text matching methods, by using the method as its underlying matching model.Four popular text matching methods have been exploited in the paper.Experimental results based on four publicly available benchmarks showed that WD-Match consistently outperformed the underlying methods and the baselines.

Most commercial industrial recommender systems have built their closed feedback loops. Though it is helpful in item recommendation and model training, the closed feedback loop may lead to the so-called bias problems, including the position bias, selection bias and popularity bias. The recommendation models trained with biased may hurt the user experiences by recommending homogenous items. How to control the biases in the closed feedback loop has become one of major challenges in modern recommender systems. This talk discusses the counterfactual learning technologies for tackling the bias problem in recommendation.

In learning-to-rank for information retrieval, a ranking model is automatically learned from the data and then utilized to rank the sets of retrieved documents. Therefore, an ideal ranking model would be a mapping from a document set to a permutation on the set, and should satisfy two critical requirements: (1) it should have the ability to model cross-document interactions so as to capture local context information in a query; (2) it should be permutation-invariant, which means that any permutation of the inputted documents would not change the output ranking. Previous studies on learning-to-rank either design uni-variate scoring functions that score each document separately, and thus failed to model the cross-document interactions; or construct multivariate scoring functions that score documents sequentially, which inevitably sacrifice the permutation invariance requirement. In this paper, we propose a neural learning-to-rank model called SetRank which directly learns a permutation-invariant ranking model defined on document sets of any size. SetRank employs a stack of (induced) multi-head self attention blocks as its key component for learning the embeddings for all of the retrieved documents jointly. The self-attention mechanism not only helps SetRank to capture the local context information from cross-document interactions, but also to learn permutation-equivariant representations for the inputted documents, which therefore achieving a permutation-invariant ranking model. Experimental results on three benchmarks showed that the SetRank significantly outperformed the baselines include the traditional learning-to-rank models and state-of-the-art Neural IR models.

This paper concerns reinforcement learning~(RL) of the document ranking models for information retrieval~(IR). One branch of the RL approaches to ranking formalize the process of ranking with Markov decision process~(MDP) and determine the model parameters with policy gradient. Though preliminary success has been shown, these approaches are still far from achieving their full potentials. Existing policy gradient methods directly utilize the absolute performance scores (returns) of the sampled document lists in its gradient estimations, which may cause two limitations: 1) fail to reflect the relative goodness of documents within the same query, which usually is close to the nature of IR ranking; 2) generate high variance gradient estimations, resulting in slow learning speed and low ranking accuracy. To deal with the issues, we propose a novel policy gradient algorithm in which the gradients are determined using pairwise comparisons of two document lists sampled within the same query. The algorithm, referred to as Pairwise Policy Gradient (PPG), repeatedly samples pairs of document lists, estimates the gradients with pairwise comparisons, and finally updates the model parameters. Theoretical analysis shows that PPG makes an unbiased and low variance gradient estimations. Experimental results have demonstrated performance gains over the state-of-the-art baselines in search result diversification and text retrieval.

Personalized search improves generic ranking models by taking user interests into consideration and returning more accurate search results to individual users. In recent years, machine learning and deep learning techniques have been successfully applied in personalized search. Most existing personalization models simply regard the search history as a static set of user behaviours and learn fixed ranking strategies based on the recorded data. Though improvements have been observed, it is obvious that these methods ignore the dynamic nature of the search process: search is a sequence of interactions between the search engine and the user. During the search process, the user interests may dynamically change. It would be more helpful if a personalized search model could track the whole interaction process and update its ranking strategy continuously. In this paper, we propose a reinforcement learning based personalization model, referred to as RLPer, to track the sequential interactions between the users and search engine with a hierarchical Markov Decision Process (MDP). In RLPer, the search engine interacts with the user to update the underlying ranking model continuously with real-time feedback. And we design a feedback-aware personalized ranking component to catch the user’s feedback which has impacts on the user interest profile for the next query. Experimental results on the publicly available AOL search log verify that our proposed model can significantly outperform state-of-the-art personalized search models.

Machine reading comprehension, whose goal is to find answers from the candidate passages for a given question, has attracted a lot of research efforts in recent years. One of the key challenge in machine reading comprehension is how to identify the main content from a large, redundant, and overlapping set of candidate sentences. In this paper we propose to tackle the challenge with Markov Decision Process in which the main content identification is formalized as sequential decision making and each action corresponds to selecting a sentence. Policy gradient is used to learn the model parameters. Experimental results based on MSMARCO showed that the proposed model, called MC-MDP, can select high quality main contents and significantly improved the performances of answer span prediction.

This paper is concerned with open-domain question answering (i.e., OpenQA). Recently, some works have viewed this problem as a reading comprehension (RC) task, and directly applied successful RC models to it. However, the performances of such models are not so good as that in the RC task. In our opinion, the perspective of RC ignores three characteristics in OpenQA task: 1) many paragraphs without the answer span are included in the data collection; 2) multiple answer spans may exist within one given paragraph; 3) the end position of an answer span is dependent with the start position. In this paper, we first propose a new probabilistic formulation of OpenQA, based on a three-level hierarchical structure, i.e., the question level, the paragraph level and the answer span level. Then a Hierarchical Answer Spans Model (HASQA) is designed to capture each probability. HAS-QA has the ability to tackle the above three problems, and experiments on public OpenQA datasets show that it significantly outperforms traditional RC baselines and recent OpenQA baselines.

Result ranking is one of the major concerns for Web search technologies. Most existing methodologies rank search results in descending order according to pointwise relevance estimation of single results. However, the dependency relationship between different search results are not taken into account. While search engine result pages contain more and more heterogenous components, a better ranking strategy should be a context-aware process and optimize result ranking globally. In this paper, we propose a novel framework which aims to improve context-aware listwise ranking performance by optimizing online evaluation metrics. The ranking problem is formalized as a Markov Decision Process (MDP) and solved with the reinforcement learning paradigm. To avoid the great cost to online systems during the training of the ranking model, we construct a virtual environment with millions of historical click logs to simulate the behavior of real users. Extensive experiments on both simulated and real datasets show that: 1) constructing a virtual environment can effectively leverage the large scale click logs and capture some important properties of real users. 2) the proposed framework can improve search ranking performance by a large margin.

In this paper we propose a novel reinforcement learning based model for named entity recognition (NER), referred to as MM-NER. Inspired by the methodology of the AlphaGo Zero, MM-NER formalizes the problem of named entity recognition with a Monte-Carlo tree search (MCTS) enhanced Markov decision process (MDP) model, in which the time steps correspond to the positions of words in a sentence from left to right, and each action corresponds to assign an NER tag to a word. Two Gated Recurrent Units (GRU) are used to summarize the past tag assignments and words in the sentence. Based on the outputs of GRUs, the policy for guiding the tag assignment and the value for predicting the whole tagging accuracy of the whole sentence are produced. The policy and value are then strengthened with MCTS, which takes the produced raw policy and value as inputs, simulates and evaluates the possible tag assignments at the subsequent positions, and outputs a better search policy for assigning tags. A reinforcement learning algorithm is proposed to train the model parameters. Empirically, we show that MM-NER can accurately predict the tags thanks to the exploratory decision making mechanism introduced by MCTS. It outperformed the conventional sequence tagging baselines and performed equally well with the state-of-the-art baseline BLSTM-CRF.

Personalized search aims to adapt document ranking to user's personal interests. Traditionally, this is done by extracting click and topical features from historical data in order to construct a user profile. In recent years, deep learning has been successfully used in personalized search due to its ability of automatic feature learning. However, the small amount of noisy personal data poses challenges to deep learning models to learn the personalized classification boundary between relevant and irrelevant results. In this paper, we propose PSGAN, a Generative Adversarial Network (GAN) framework for personalized search. By means of adversarial training, we enforce the model to pay more attention to training data that are difficult to distinguish. We use the discriminator to evaluate personalized relevance of documents and use the generator to learn the distribution of relevant documents. Two alternative ways to construct the generator in the framework are tested: based on the current query or based on a set of generated queries. Experiments on data from a commercial search engine show that our models can yield significant improvements over state-of-the-art models.

Matching is the key problem in search and recommendation, that is to measure the relevance of a document to a query or the interest of a user on an item. Previously, machine learning methods have been exploited to address the problem, which learn a matching function from labeled data, also referred to as "learning to match". In recent years, deep learning has been successfully applied to matching and significant progresses have been made. Deep semantic matching models for search and neural collaborative filtering models for recommendation are becoming the state-of-the-art technologies. The key to the success of the deep learning approach is its strong ability in learning of representations and generalization of matching patterns from raw data (e.g., queries, documents, users, and items, particularly in their raw forms). In this tutorial, we aim to give a comprehensive survey on recent progress in deep learning for matching in search and recommendation. Our tutorial is unique in that we try to give a unified view on search and recommendation. In this way, we expect researchers from the two fields can get deep understanding and accurate insight on the spaces, stimulate more ideas and discussions, and promote developments of technologies. The tutorial mainly consists of three parts. Firstly, we introduce the general problem of matching, which is fundamental in both search and recommendation. Secondly, we explain how traditional machine learning techniques are utilized to address the matching problems in search and recommendation. Lastly, we elaborate how deep learning can be effectively used to solve the matching problems in both tasks.

Ranking SVM, which formalizes the problem of learning a ranking model as that of learning a binary SVM on preference pairs of documents, is a state-of-the-art ranking model in information retrieval. The dual form solution of a linear Ranking SVM model can be written as a linear combination of the preference pairs, i.e., w = Σ(i,j)αijxi-xj), where αijdenotes the Lagrange parameters associated with each preference pair (i,j). It is observed that there exist obvious interactions among the document pairs because two preference pairs could share a same document as their items, e.g., preference pairs (d1,d2) and (d1,d3) share the document d1. Thus it is natural to ask if there also exist interactions over the model parameters αij, which may be leveraged to construct better ranking models. This paper aims to answer the question. We empirically found that there exists a low-rank structure over the rearranged Ranking SVM model parameters αij, which indicates that the interactions do exist. Based on the discovery, we made modifications on the original Ranking SVM model by explicitly applying low-rank constraints to the Lagrange parameters, achieving two novel algorithms called Factorized Ranking SVM and Regularized Ranking SVM, respectively. Specifically, in Factorized Ranking SVM each parameter αijis decomposed as a product of two low-dimensional vectors, i.e., αij=〈vi,vj〉, where vectors viand vjcorrespond to document i and j, respectively; In Regularized Ranking SVM, a nuclear norm is applied to the rearranged parameters matrix for controlling its rank. Experimental results on three LETOR datasets show that both of the proposed methods can outperform state-of-the-art learning to rank models including the conventional Ranking SVM.

Approximate nearest neighbor search is a fundamental problem and has been studied for a few decades. Recently graph-based indexing methods have demonstrated their great efficiency, whose main idea is to construct neighborhood graph offline and perform a greedy search starting from some sampled points of the graph online. Most existing graph-based methods focus on either the precise k-nearest neighbor (k-NN) graph which has good exploitation ability, or the diverse graph which has good exploration ability. In this paper, we propose the k-diverse nearest neighbor (k-DNN) graph, which balances the precision and diversity of the graph, leading to good exploitation and exploration abilities simultaneously. We introduce an efficient indexing algorithm for the construction of the k-DNN graph inspired by a well-known diverse ranking algorithm in information retrieval (IR). Experimental results show that our method can outperform both state-of-the-art precise graph and diverse graph methods.

Sequence to sequence (Seq2Seq) models have been widely used for response generation in the area of conversation.However, the requirements for different conversation scenarios are distinct.For example, customer service requires the generated responses to be specific and accurate, while chatbot prefers diverse responses so as to attract different users.The current Seq2Seq model fails to meet these diverse requirements, by using a general average likelihood as the optimization criteria.As a result, it usually generates safe and commonplace responses, such as 'I don't know'.In this paper, we propose two tailored optimization criteria for Seq2Seq to different conversation scenarios, i.e., the maximum generated likelihood for specific-requirement scenario, and the conditional value-at-risk for diverse-requirement scenario.Experimental results on the Ubuntu dialogue corpus (Ubuntu service scenario) and Chinese Weibo dataset (social chatbot scenario) show that our proposed models not only satisfies diverse requirements for different scenarios, but also yields better performances against traditional Seq2Seq models in terms of both metric-based and human evaluations.

In conversation, a general response (e.g., "I don't know") could correspond to a large variety of input utterances.Previous generative conversational models usually employ a single model to learn the relationship between different utteranceresponse pairs, thus tend to favor general and trivial responses which appear frequently.To address this problem, we propose a novel controlled response generation mechanism to handle different utterance-response relationships in terms of specificity.Specifically, we introduce an explicit specificity control variable into a sequence-to-sequence model, which interacts with the usage representation of words through a Gaussian Kernel layer, to guide the model to generate responses at different specificity levels.We describe two ways to acquire distant labels for the specificity control variable in learning.Empirical studies show that our model can significantly outperform the state-of-theart response generation models under both automatic and human evaluations.

Hashing techniques for approximate nearest neighbor search (ANNS) encode data points into a set of short binary codes, while trying to preserve the neighborhood structure of the original data as much as possible. With the binary codes, the task of ANNS can be easily conducted over large-scale dataset, due to the high efficiency of pairwise comparison with the Hamming distance. Although binary codes have low computation and storage cost, the data are heavily compressed so that partial neighborhood structure information would be inevitably lost. To address this issue, we propose to introduce the k-nearest neighbors (k-NNs) in the original space into the Hamming space (i.e., associating a binary code with its original k-NNs) to enhance the effectiveness of existing hashing techniques with little overhead. Based on this idea, we develop a novel search scheme for hashing techniques namely neighborhood voting, i.e., each point retrieved by a query code will vote for its neighbors and itself, and the more voted, the better candidates. In this way, search in hashing is not simply the collision between codes (i.e., query code and candidate code), but also the collision between neighbors (i.e., neighbors of candidate points). The underlying assumption is that the true neighbors of a query point should be close to each other, while points with similar binary codes but seldom be the neighbors of other candidate points would be false positives. We introduce a novel data structure called aggregated hash table for implementing our idea and accelerating the online search process. Experimental results show that our search scheme can significantly improve the search effectiveness while having good efficiency over different existing hashing techniques.

In this paper, we introduce and tackle the Question Headline Generation (QHG) task. The motivation comes from the investigation of a real-world news portal where we find that news articles with question headlines often receive much higher click-through ratio than those with non-question headlines. The QHG task can be viewed as a specific form of the Question Generation (QG) task, with the emphasis on creating a natural question from a given news article by taking the entire article as the answer. A good QHG model thus should be able to generate a question by summarizing the essential topics of an article. Based on this idea, we propose a novel dual-attention sequence-to-sequence model (DASeq2Seq) for the QHG task. Unlike traditional sequence-to-sequence models which only employ the attention mechanism in the decoding phase for better generation, our DASeq2Seq further introduces a self-attention mechanism in the encoding phase to help generate a good summary of the article. We investigate two ways of the self-attention mechanism, namely global self-attention and distributed self-attention. Besides, we employ a vocabulary gate over both generic and question vocabularies to better capture the question patterns. Through the offline experiments, we show that our approach can significantly outperform the state-of-the-art question generation or headline generation models. Furthermore, we also conduct online evaluation to demonstrate the effectiveness of our approach using A/B test.

Sequence to sequence (Seq2Seq) approach has gained great attention in the field of single-turn dialogue generation. However, one serious problem is that most existing Seq2Seq based models tend to generate common responses lacking specific meanings. Our analysis show that the underlying reason is that Seq2Seq is equivalent to optimizing Kullback–Leibler (KL) divergence, thus does not penalize the case whose generated probability is high while the true probability is low. However, the true probability is unknown, which poses challenges for tackling this problem. Inspired by the fact that the coherence (i.e. similarity) between post and response is consistent with human evaluation, we hypothesize that the true probability of a response is proportional to the coherence degree. The coherence scores are then used as the reward function in a reinforcement learning framework to penalize the case whose generated probability is high while the true probability is low. Three different types of coherence models, including an unlearned similarity function, a pretrained semantic matching function, and an end-to-end dual learning architecture, are proposed in this paper. Experimental results on both Chinese Weibo dataset and English Subtitle dataset show that the proposed models produce more specific and meaningful responses, yielding better performances against Seq2Seq models in terms of both metric-based and human evaluations.

Assessing relevance between a query and a document is challenging in ad-hoc retrieval due to its diverse patterns, i.e., a document could be relevant to a query as a whole or partially as long as it provides sufficient information for users' need. Such diverse relevance patterns require an ideal retrieval model to be able to assess relevance in the right granularity adaptively. Unfortunately, most existing retrieval models compute relevance at a single granularity, either document-wide or passage-level, or use fixed combination strategy, restricting their ability in capturing diverse relevance patterns. In this work, we propose a data-driven method to allow relevance signals at different granularities to compete with each other for final relevance assessment. Specifically, we propose a HIerarchical Neural maTching model (HiNT) which consists of two stacked components, namely local matching layer and global decision layer. The local matching layer focuses on producing a set of local relevance signals by modeling the semantic matching between a query and each passage of a document. The global decision layer accumulates local signals into different granularities and allows them to compete with each other to decide the final relevance score.Experimental results demonstrate that our HiNT model outperforms existing state-of-the-art retrieval models significantly on benchmark ad-hoc retrieval datasets.

The goal of search result diversification is to select a subset of documents from the candidate set to satisfy as many different subtopics as possible. In general, it is a problem of subset selection and selecting an optimal subset of documents is NP-hard. Existing methods usually formalize the problem as ranking the documents with greedy sequential document selection. At each of the ranking position the document that can provide the largest amount of additional information is selected. It is obvious that the greedy selections inevitably produce suboptimal rankings. In this paper we propose to partially alleviate the problem with a Monte Carlo tree search (MCTS) enhanced Markov decision process (MDP), referred to as M$^2$Div. In M$^2$Div, the construction of diverse ranking is formalized as an MDP process where each action corresponds to selecting a document for one ranking position. Given an MDP state which consists of the query, selected documents, and candidates, a recurrent neural network is utilized to produce the policy function for guiding the document selection and the value function for predicting the whole ranking quality. The produced raw policy and value are then strengthened with MCTS through exploring the possible rankings at the subsequent positions, achieving a better search policy for decision-making. Experimental results based on the TREC benchmarks showed that M$^2$Div can significantly outperform the state-of-the-art baselines based on greedy sequential document selection, indicating the effectiveness of the exploratory decision-making mechanism in M$^2$Div.

Matching is the key problem in both search and recommendation, that is to measure the relevance of a document to a query or the interest of a user on an item. Previously, machine learning methods have been exploited to address the problem, which learns a matching function from labeled data, also referred to as "learning to match''. In recent years, deep learning has been successfully applied to matching and significant progresses have been made. Deep semantic matching models for search and neural collaborative filtering models for recommendation are becoming the state-of-the-art technologies. The key to the success of the deep learning approach is its strong ability in learning of representations and generalization of matching patterns from raw data (e.g., queries, documents, users, and items, particularly in their raw forms). In this tutorial, we aim to give a comprehensive survey on recent progress in deep learning for matching in search and recommendation. Our tutorial is unique in that we try to give a unified view on search and recommendation. In this way, we expect researchers from the two fields can get deep understanding and accurate insight on the spaces, stimulate more ideas and discussions, and promote developments of technologies. The tutorial mainly consists of three parts. Firstly, we introduce the general problem of matching, which is fundamental in both search and recommendation. Secondly, we explain how traditional machine learning techniques are utilized to address the matching problem in search and recommendation. Lastly, we elaborate how deep learning can be effectively used to solve the matching problems in both tasks.

Policy gradient, which makes use of Monte Carlo method to get an unbiased estimation of the parameter gradients, has been widely used in reinforcement learning. One key issue in policy gradient is reducing the variance of the estimation. From the viewpoint of statistics, policy gradient with baseline, a successful variance reduction method for policy gradient, directly applies the control variates method, a traditional variance reduction technique used in Monte Carlo, to policy gradient. One problem with control variates method is that the quality of estimation heavily depends on the choice of the control variates. To address the issue and inspired by the antithetic variates method for variance reduction, we propose to combine the antithetic variates method with traditional policy gradient for the multi-armed bandit problem. Furthermore, we achieve a new policy gradient algorithm called Antithetic-Arm Bandit (AAB). In AAB, the gradient is estimated through coordinate ascent where at each iteration gradient of the target arm is estimated through: 1) constructing a sequence of arms which is approximately monotonic in terms of estimated gradients, 2) sampling a pair of antithetic arms over the sequence, and 3) re-estimating the target gradient based on the sampled pair. Theoretical analysis proved that AAB achieved an unbiased and variance reduced estimation. Experimental results based on a multi-armed bandit task showed that AAB can achieve state-of-the-art performances.

Convolutional Neural Networks (CNN) and the locally connected layer are limited in capturing the importance and relations of different local receptive fields, which are often crucial for tasks such as face verification, visual question answering, and word sequence prediction. To tackle the issue, we propose a novel locally smoothed neural network (LSNN) in this paper. The main idea is to represent the weight matrix of the locally connected layer as the product of the kernel and the smoother, where the kernel is shared over different local receptive fields, and the smoother is for determining the importance and relations of different local receptive fields. Specifically, a multi-variate Gaussian function is utilized to generate the smoother, for modeling the location relations among different local receptive fields. Furthermore, the content information can also be leveraged by setting the mean and precision of the Gaussian function according to the content. Experiments on some variant of MNIST clearly show our advantages over CNN and locally connected layer.

When applying learning to rank algorithms to Web search, a large number of features are usually designed to capture the relevance signals. Most of these features are computed based on the extracted textual elements, link analysis, and user logs. However, Web pages are not solely linked texts, but have structured layout organizing a large variety of elements in different styles. Such layout itself can convey useful visual information, indicating the relevance of a Web page. For example, the query-independent layout (i.e., raw page layout) can help identify the page quality, while the query-dependent layout (i.e., page rendered with matched query words) can further tell rich structural information (e.g., size, position and proximity) of the matching signals. However, such visual information of layout has been seldom utilized in Web search in the past. In this work, we propose to learn rich visual features automatically from the layout of Web pages (i.e., Web page snapshots) for relevance ranking. Both query-independent and query-dependent snapshots are considered as the new inputs. We then propose a novel visual perception model inspired by human's visual search behaviors on page viewing to extract the visual features. This model can be learned end-to-end together with traditional human-crafted features. We also show that such visual features can be efficiently acquired in the online setting with an extended inverted indexing scheme. Experiments on benchmark collections demonstrate that learning visual features from Web page snapshots can significantly improve the performance of relevance ranking in ad-hoc Web retrieval tasks.

This paper concerns a deep learning approach to relevance ranking in information retrieval (IR). Existing deep IR models such as DSSM and CDSSM directly apply neural networks to generate ranking scores, without explicit understandings of the relevance. According to the human judgement process, a relevance label is generated by the following three steps: 1) relevant locations are detected; 2) local relevances are determined; 3) local relevances are aggregated to output the relevance label. In this paper we propose a new deep learning architecture, namely DeepRank, to simulate the above human judgment process. Firstly, a detection strategy is designed to extract the relevant contexts. Then, a measure network is applied to determine the local relevances by utilizing a convolutional neural network (CNN) or two-dimensional gated recurrent units (2D-GRU). Finally, an aggregation network with sequential integration and term gating mechanism is used to produce a global relevance score. DeepRank well captures important IR characteristics, including exact/semantic matching signals, proximity heuristics, query term importance, and diverse relevance requirement. Experiments on both benchmark LETOR dataset and a large scale clickthrough data show that DeepRank can significantly outperform learning to ranking methods, and existing deep learning methods.

One of the central issues in learning to rank for information retrieval is to develop algorithms that construct ranking models by directly optimizing evaluation measures such as normalized discounted cumulative gain~(ND CG). Existing methods usually focus on optimizing a specific evaluation measure calculated at a fixed position, e.g., NDCG calculated at a fixed position K. In information retrieval the evaluation measures, including the widely used NDCG and [email protected], are usually designed to evaluate the document ranking at all of the ranking positions, which provide much richer information than only measuring the document ranking at a single position. Thus, it is interesting to ask if we can devise an algorithm that has the ability of leveraging the measures calculated at all of the ranking postilions, for learning a better ranking model. In this paper, we propose a novel learning to rank model on the basis of Markov decision process (MDP), referred to as MDPRank. In the learning phase of MDPRank, the construction of a document ranking is considered as a sequential decision making, each corresponds to an action of selecting a document for the corresponding position. The policy gradient algorithm of REINFORCE is adopted to train the model parameters. The evaluation measures calculated at every ranking positions are utilized as the immediate rewards to the corresponding actions, which guide the learning algorithm to adjust the model parameters so that the measure is optimized. Experimental results on LETOR benchmark datasets showed that MDPRank can outperform the state-of-the-art baselines.

In this paper we address the issue of learning diverse ranking models for search result diversification. Typical methods treat the problem of constructing a diverse ranking as a process of sequential document selection. At each ranking position, the document that can provide the largest amount of additional information to the users is selected, because the search users usually browse the documents in a top-down manner. Thus, to select an optimal document for a position, it is critical for a diverse ranking model to capture the utility of information the user have perceived from the preceding documents. Existing methods usually calculate the ranking scores (e.g., the marginal relevance) directly based on the query and the selected documents, with heuristic rules or handcrafted features. The utility the user perceived at each of the ranks, however, is not explicitly modeled. In this paper, we present a novel diverse ranking model on the basis of continuous state Markov decision process (MDP) in which the user perceived utility is modeled as a part of the MDP state. Our model, referred to as MDP-DIV, sequentially takes the actions of selecting one document according to current state, and then updates the state for the chosen of the next action. The transition of the states are modeled in a recurrent manner and the model parameters are learned with policy gradient. Experimental results based on the TREC benchmarks showed that MDP-DIV can significantly outperform the state-of-the-art baselines.

The queries issued to search engines are often ambiguous or multifaceted, which requires search engines to return diverse results that can fulfill as many different information needs as possible; this is called search result diversification . Recently, the relational learning to rank model, which designs a learnable ranking function following the criterion of maximal marginal relevance, has shown effectiveness in search result diversification [Zhu et al. 2014]. The goodness of a diverse ranking model is usually evaluated with diversity evaluation measures such as α-NDCG [Clarke et al. 2008], ERR-IA [Chapelle et al. 2009], and D#-NDCG [Sakai and Song 2011]. Ideally the learning algorithm would train a ranking model that could directly optimize the diversity evaluation measures with respect to the training data. Existing relational learning to rank algorithms, however, only train the ranking models by optimizing loss functions that loosely relate to the evaluation measures. To deal with the problem, we propose a general framework for learning relational ranking models via directly optimizing any diversity evaluation measure . In learning, the loss function upper-bounding the basic loss function defined on a diverse ranking measure is minimized. We can derive new diverse ranking algorithms under the framework, and several diverse ranking algorithms are created based on different upper bounds over the basic loss function. We conducted comparisons between the proposed algorithms with conventional diverse ranking methods using the TREC benchmark datasets. Experimental results show that the algorithms derived under the diverse learning to rank framework always significantly outperform the state-of-the-art baselines.

Matching a question to its best answer is a common task in community question answering. In this paper, we focus on the non-factoid questions and aim to pick out the best answer from its candidate answers. Most of the existing deep models directly measure the similarity between question and answer by their individual sentence embeddings. In order to tackle the problem of the information lack in question's descriptions and the lexical gap between questions and answers, we propose a novel deep architecture namely SPAN in this paper. Specifically we introduce support answers to help understand the question, which are defined as the best answers of those similar questions to the original one. Then we can obtain two kinds of similarities, one is between question and the candidate answer, and the other one is between support answers and the candidate answer. The matching score is finally generated by combining them. Experiments on Yahoo! Answers demonstrate that SPAN can outperform the baseline models.

Matching two texts is a fundamental problem in many natural language processing tasks. An effective way is to extract meaningful matching patterns from words, phrases, and sentences to produce the matching score. Inspired by the success of convolutional neural network in image recognition, where neurons can capture many complicated patterns based on the extracted elementary visual patterns such as oriented edges and corners, we propose to model text matching as the problem of image recognition. Firstly, a matching matrix whose entries represent the similarities between words is constructed and viewed as an image. Then a convolutional neural network is utilized to capture rich matching patterns in a layer-by-layer way. We show that by resembling the compositional hierarchies of patterns in image recognition, our model can successfully identify salient signals such as n-gram and n-term matchings. Experimental results demonstrate its superiority against the baselines.

Distributional hypothesis lies in the root of most existing word representation models by inferring word meaning from its external contexts. However, distributional models cannot handle rare and morphologically complex words very well and fail to identify some fine-grained linguistic regularity as they are ignoring the word forms. On the contrary, morphology points out that words are built from some basic units, i.e., morphemes. Therefore, the meaning and function of such rare words can be inferred from the words sharing the same morphemes, and many syntactic relations can be directly identified based on the word forms. However, the limitation of morphology is that it cannot infer the relationship between two words that do not share any morphemes. Considering the advantages and limitations of both approaches, we propose two novel models to build better word representations by modeling both external contexts and internal morphemes in a jointly predictive way, called BEING and SEING. These two models can also be extended to learn phrase representations according to the distributed morphology theory. We evaluate the proposed models on similarity tasks and analogy tasks. The results demonstrate that the proposed models can outperform state-of-the-art models significantly on both word and phrase representation learning.

Matching natural language sentences is central for many applications such as information retrieval and question answering. Existing deep models rely on a single sentence representation or multiple granularity representations for matching. However, such methods cannot well capture the contextualized local information in the matching process. To tackle this problem, we present a new deep architecture to match two sentences with multiple positional sentence representations. Specifically, each positional sentence representation is a sentence representation at this position, generated by a bidirectional long short term memory (Bi-LSTM). The matching score is finally produced by aggregating interactions between these different positional sentence representations, through k-Max pooling and a multi-layer perceptron. Our model has several advantages: (1) By using Bi-LSTM, rich context of the whole sentence is leveraged to capture the contextualized local information in each positional sentence representation; (2) By matching with multiple positional sentence representations, it is flexible to aggregate different important contextualized local information in a sentence to support the matching; (3) Experiments on different tasks such as question answering and sentence completion demonstrate the superiority of our model.

Machine learning algorithms have become the key components in many big data applications. However, the full potential of machine learning is still far from been realized because using machine learning algorithms is hard, especially on distributed platforms such as Hadoop and Spark. The key barriers come from not only the implementation of the algorithms themselves, but also the processing for applying them to real applications which often involve multiple steps and different algorithms. In this demo we present a general-purpose dataflow-based system for easing the process of applying machine learning algorithms to real world tasks. In the system, a learning task is formulated as a directed acyclic graph (DAG) in which each node represents an operation (e.g., a machine learning algorithm), and each edge represents the flow of the data from one node to its descendants. Graphical user interface is implemented for making users to create, configure, submit, and monitor a task in a drag-and-drop manner. Advantages of the system include 1) lowering the barriers of defining and executing machine learning tasks; 2) sharing and re-using the implementations of the algorithms, the task dataflow DAGs, and the (intermediate) experimental results; 3) seamlessly integrating the stand-alone algorithms as well as the distributed algorithms in one task. The system has been deployed as a machine learning service and can be access from the Internet.

Search result diversification has attracted considerable attention as a means to tackle the ambiguous or multi-faceted information needs of users. One of the key problems in search result diversification is novelty, that is, how to measure the novelty of a candidate document with respect to other documents. In the heuristic approaches, the predefined document similarity functions are directly utilized for defining the novelty. In the learning approaches, the novelty is characterized based on a set of handcrafted features. Both the similarity functions and the features are difficult to manually design in real world due to the complexity of modeling the document novelty. In this paper, we propose to model the novelty of a document with a neural tensor network. Instead of manually defining the similarity functions or features, the new method automatically learns a nonlinear novelty function based on the preliminary representation of the candidate document and other documents. New diverse learning to rank models can be derived under the relational learning to rank framework. To determine the model parameters, loss functions are constructed and optimized with stochastic gradient descent. Extensive experiments on three public TREC datasets show that the new derived algorithms can significantly outperform the baselines, including the state-of-the-art relational learning to rank models.

Demographic attributes play an important role in retail market to characterize different types of users. Such signals however are often only available for a small fraction of users in practice due to the difficulty in manual collection process by retailers. In this paper, we aim to harness the power of big data to automatically infer users' demographic attributes based on their purchase data. Typically, demographic prediction can be formalized as a multi-task multi-class prediction problem, i.e., multiple demographic attributes (e.g., gender, age and income) are to be inferred for each user where each attribute may belong to one of N possible classes (N-2). Most previous work on this problem explores different types of features and usually predicts different attributes independently. However, modeling the tasks separately may lose the ability to leverage the correlations among different attributes. Meanwhile, manually defined features require professional knowledge and often suffer from under specification. To address these problems, we propose a novel Structured Neural Embedding (SNE) model to automatically learn the representations from users' purchase data for predicting multiple demographic attributes simultaneously. Experiments are conducted on a real-world retail dataset where five attributes (gender, marital status, income, age, and education level) are to be predicted. The empirical results show that our SNE model can improve the performance significantly compared with state-of-the-art baselines.

Bursty topics discovery in microblogs is important for people to grasp essential and valuable information. However, the task is challenging since microblog posts are particularly short and noisy. This work develops a novel probabilistic model, namely Bursty Biterm Topic Model (BBTM), to deal with the task. BBTM extends the Biterm Topic Model (BTM) by incorporating the burstiness of biterms as prior knowledge for bursty topic modeling, which enjoys the following merits: 1) It can well solve the data sparsity problem in topic modeling over short texts as the same as BTM; 2) It can automatical discover high quality bursty topics in microblogs in a principled and efficient way. Extensive experiments on a standard Twitter dataset show that our approach outperforms the state-of-the-art baselines significantly.

Fei Sun, Jiafeng Guo, Yanyan Lan, Jun Xu, Xueqi Cheng. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2015.

Ranking SVM, which formalizes the problem of learning a ranking model as that of learning a binary SVM on preference pairs of documents, is a state-of-the-art ranking model in information retrieval. The dual form solution of Ranking SVM model can be written as a linear combination of the preference pairs, i.e., w = ∑(i,j) αij (xi - xj), where αij denotes the Lagrange parameters associated with each pair (i,j). It is obvious that there exist significant interactions over the document pairs because two preference pairs could share a same document as their items. Thus it is natural to ask if there also exist interactions over the model parameters αij, which we may leverage to propose better ranking model. This paper aims to answer the question. Firstly, we found that there exists a low-rank structure over the Ranking SVM model parameters αij, which indicates that the interactions do exist. Then, based on the discovery, we made a modification on the original Ranking SVM model by explicitly applying a low-rank constraint to the parameters. Specifically, each parameter αij is decomposed as a product of two low-dimensional vectors, i.e., αij = vi, vj, where vectors vi and vj correspond to document i and j, respectively. The learning process, thus, becomes to optimize the modified dual form objective function with respect to the low-dimensional vectors. Experimental results on three LETOR datasets show that our method, referred to as Factorized Ranking SVM, can outperform state-of-the-art baselines including the conventional Ranking SVM.

Next basket recommendation is a crucial task in market basket analysis. Given a user's purchase history, usually a sequence of transaction data, one attempts to build a recommender that can predict the next few items that the user most probably would like. Ideally, a good recommender should be able to explore the sequential behavior (i.e., buying one item leads to buying another next), as well as account for users' general taste (i.e., what items a user is typically interested in) for recommendation. Moreover, these two factors may interact with each other to influence users' next purchase. To tackle the above problems, in this paper, we introduce a novel recommendation approach, namely hierarchical representation model (HRM). HRM can well capture both sequential behavior and users' general taste by involving transaction and user representations in prediction. Meanwhile, the flexibility of applying different aggregation operations, especially nonlinear operations, on representations allows us to model complicated interactions among different factors. Theoretically, we show that our model subsumes several existing methods when choosing proper aggregation operations. Empirically, we demonstrate that our model can consistently outperform the state-of-the-art baselines under different evaluation metrics on real-world transaction data.

In this paper we address the issue of learning a ranking model for search result diversification. In the task, a model concerns with both query-document relevance and document diversity is automatically created with training data. Ideally a diverse ranking model would be designed to meet the criterion of maximal marginal relevance, for selecting documents that have the least similarity to previously selected documents. Also, an ideal learning algorithm for diverse ranking would train a ranking model that could directly optimize the diversity evaluation measures with respect to the training data. Existing methods, however, either fail to model the marginal relevance, or train ranking models by minimizing loss functions that loosely related to the evaluation measures. To deal with the problem, we propose a novel learning algorithm under the framework of Perceptron, which adopts the ranking model that \emph{maximizes marginal relevance at ranking and can optimize any diversity evaluation measure in training}. The algorithm, referred to as PAMM (Perceptron Algorithm using Measures as Margins), first constructs positive and negative diverse rankings for each training query, and then repeatedly adjusts the model parameters so that the margins between the positive and negative rankings are maximized. Experimental results on three benchmark datasets show that PAMM significantly outperforms the state-of-the-art baseline methods.

This paper addresses the problem of post-processing of ranking in search, referred to as post ranking. Although important, no research seems to have been conducted on the problem, particularly with a principled approach, and in practice ad-hoc ways of performing the task are being adopted. This paper formalizes the problem as constrained optimization in which the constraints represent the post-processing rules and the objective function represents the trade-off between adherence to the original ranking and satisfaction of the rules. The optimization amounts to refining the original ranking result based on the rules. We further propose a specific probabilistic implementation of the general formalization on the basis of the Bradley-Terry model, which is theoretically sound, effective, and efficient. Our experimental results, using benchmark datasets and enterprise search dataset, show that the proposed method works much better than several baseline methods of utilizing rules.

Recently, significant progress has been made in research on what we call semantic matching (SM), in web search, question answering, online advertisement, cross-language information retrieval, and other tasks. Advanced technologies based on machine learning have been developed. Let us take Web search as example of the problem that also pervades the other tasks. When comparing the textual content of query and documents, Web search still heavily relies on the term-based approach, where the relevance scores between queries and documents are calculated on the basis of the degree of matching between query terms and document terms. This simple approach works rather well in practice, partly because there are many other signals in web search (hypertext, user logs, etc.) that complement it. However, when considering the long tail of web searches, it can suffer from data sparseness, e.g., Trenton does not match New Jersey Capital. Query document mismatches occur when searcher and author use different terms (representations), and this phenomenon is prevalent due to the nature of human language.

We consider learning query and document similarities from a click-through bipartite graph with metadata on the nodes. The metadata contains multiple types of features of queries and documents. We aim to leverage both the click-through bipartite graph and the features to learn query-document, document-document, and query-query similarities. The challenges include how to model and learn the similarity functions based on the graph data.

Topic modeling provides a powerful way to analyze the content of a collection of documents. It has become a popular tool in many research areas, such as text mining, information retrieval, natural language processing, and other related fields. In real-world applications, however, the usefulness of topic modeling is limited due to scalability issues. Scaling to larger document collections via parallelization is an active area of research, but most solutions require drastic steps, such as vastly reducing input vocabulary. In this article we introduce Regularized Latent Semantic Indexing (RLSI)---including a batch version and an online version, referred to as batch RLSI and online RLSI, respectively---to scale up topic modeling. Batch RLSI and online RLSI are as effective as existing topic modeling techniques and can scale to larger datasets without reducing input vocabulary. Moreover, online RLSI can be applied to stream data and can capture the dynamic evolution of topics. Both versions of RLSI formalize topic modeling as a problem of minimizing a quadratic loss function regularized by ℓ1 and/or ℓ2 norm. This formulation allows the learning process to be decomposed into multiple suboptimization problems which can be optimized in parallel, for example, via MapReduce. We particularly propose adopting ℓ1 norm on topics and ℓ2 norm on document representations to create a model with compact and readable topics and which is useful for retrieval. In learning, batch RLSI processes all the documents in the collection as a whole, while online RLSI processes the documents in the collection one by one. We also prove the convergence of the learning of online RLSI. Relevance ranking experiments on three TREC datasets show that batch RLSI and online RLSI perform better than LSI, PLSI, LDA, and NMF, and the improvements are sometimes statistically significant. Experiments on a Web dataset containing about 1.6 million documents and 7 million terms, demonstrate a similar boost in performance.

No abstract available.

Topic modeling can reveal the latent structure of text data and is useful for knowledge discovery, search relevance ranking, document classification, and so on. One of the major challenges in topic modeling is to deal with large datasets and large numbers of topics in real-world applications. In this paper, we investigate techniques for scaling up the non-probabilistic topic modeling approaches such as RLSI and NMF. We propose a general topic modeling method, referred to as Group Matrix Factorization (GMF), to enhance the scalability and efficiency of the non-probabilistic approaches. GMF assumes that the text documents have already been categorized into multiple semantic classes, and there exist class-specific topics for each of the classes as well as shared topics across all classes. Topic modeling is then formalized as a problem of minimizing a general objective function with regularizations and/or constraints on the class-specific topics and shared topics. In this way, the learning of class-specific topics can be conducted in parallel, and thus the scalability and efficiency can be greatly improved. We apply GMF to RLSI and NMF, obtaining Group RLSI (GRLSI) and Group NMF (GNMF) respectively. Experiments on a Wikipedia dataset and a real-world web dataset, each containing about 3 million documents, show that GRLSI and GNMF can greatly improve RLSI and NMF in terms of scalability and efficiency. The topics discovered by GRLSI and GNMF are coherent and have good readability. Further experiments on a search relevance dataset, containing 30,000 labeled queries, show that the use of topics learned by GRLSI and GNMF can significantly improve search relevance.

In web search, relevance is one of the most important factors to meet users' satisfaction, and the success of a web search engine heavily depends on its performance on relevance. It has been observed that many hard cases in search relevance are due to term mismatch between query and documnt (e.g., query 'ny times' does not match well with document only containing 'new york times'), and thus it is not exaggerated to say that dealing with mismatch between query and document is one of the most critical research problems in web search. Recently researchers have spent significant effort to address the grand challenge. The major approach is to conduct more query and document understanding, and perform better matching between enriched query and document representations. With the availability of large amount of log data and advanced machine learning techniques, this becomes more feasible and significant progress has been made recently.

Topic modeling can boost the performance of information retrieval, but its real-world application is limited due to scalability issues. Scaling to larger document collections via parallelization is an active area of research, but most solutions require drastic steps such as vastly reducing input vocabulary. We introduce Regularized Latent Semantic Indexing (RLSI), a new method which is designed for parallelization. It is as effective as existing topic models, and scales to larger datasets without reducing input vocabulary. RLSI formalizes topic modeling as a problem of minimizing a quadratic loss function regularized by l₂ and/or l₁ norm. This formulation allows the learning process to be decomposed into multiple sub-optimization problems which can be optimized in parallel, for example via MapReduce. We particularly propose adopting l₂ norm on topics and l₁ norm on document representations, to create a model with compact and readable topics and useful for retrieval. Relevance ranking experiments on three TREC datasets show that RLSI performs better than LSI, PLSI, and LDA, and the improvements are sometimes statistically significant. Experiments on a web dataset, containing about 1.6 million documents and 7 million terms, demonstrate a similar boost in performance on a larger corpus and vocabulary than in previous studies.

This paper points out that many machine learning problems in IR should be and can be formalized in a novel way, referred to as 'group-based learning'. In group-based learning, it is assumed that training data as well as testing data consist of groups. The classifier is created and utilized across groups. Furthermore, evaluation in testing and also in training are conducted at group level, with the use of evaluation measures defined on a group. This paper addresses the problem and presents a Boosting algorithm to perform the new learning task. The algorithm, referred to as AdaBoost.Group, is proved to be able to improve accuracies in terms of group-based measures during training.

One of the central issues in learning to rank for information retrieval is to develop algorithms that construct ranking models by directly optimizing evaluation measures used in information retrieval such as Mean Average Precision (MAP) and Normalized Discounted Cumulative Gain (NDCG). Several such algorithms including SVMmap and AdaRank have been proposed and their effectiveness has been verified. However, the relationships between the algorithms are not clear, and furthermore no comparisons have been conducted between them. In this paper, we conduct a study on the approach of directly optimizing evaluation measures in learning to rank for Information Retrieval (IR). We focus on the methods that minimize loss functions upper bounding the basic loss function defined on the IR measures. We first provide a general framework for the study and analyze the existing algorithms of SVMmap and AdaRank within the framework. The framework is based on upper bound analysis and two types of upper bounds are discussed. Moreover, we show that we can derive new algorithms on the basis of this analysis and create one example algorithm called PermuRank. We have also conducted comparisons between SVMmap, AdaRank, PermuRank, and conventional methods of Ranking SVM and RankBoost, using benchmark datasets. Experimental results show that the methods based on direct optimization of evaluation measures can always outperform conventional methods of Ranking SVM and RankBoost. However, no significant difference exists among the performances of the direct optimization methods themselves.

In this paper we address the issue of learning to rank for document retrieval. In the task, a model is automatically created with some training data and then is utilized for ranking of documents. The goodness of a model is usually evaluated with performance measures such as MAP (Mean Average Precision) and NDCG (Normalized Discounted Cumulative Gain). Ideally a learning algorithm would train a ranking model that could directly optimize the performance measures with respect to the training data. Existing methods, however, are only able to train ranking models by minimizing loss functions loosely related to the performance measures. For example, Ranking SVM and RankBoost train ranking models by minimizing classification errors on instance pairs. To deal with the problem, we propose a novel learning algorithm within the framework of boosting, which can minimize a loss function directly defined on the performance measures. Our algorithm, referred to as AdaRank, repeatedly constructs 'weak rankers' on the basis of reweighted training data and finally linearly combines the weak rankers for making ranking predictions. We prove that the training process of AdaRank is exactly that of enhancing the performance measure used. Experimental results on four benchmark datasets show that AdaRank significantly outperforms the baseline methods of BM25, Ranking SVM, and RankBoost.

The paper is concerned with applying learning to rank to document retrieval. Ranking SVM is a typical method of learning to rank. We point out that there are two factors one must consider when applying Ranking SVM, in general a "learning to rank" method, to document retrieval. First, correctly ranking documents on the top of the result list is crucial for an Information Retrieval system. One must conduct training in a way that such ranked results are accurate. Second, the number of relevant documents can vary from query to query. One must avoid training a model biased toward queries with a large number of relevant documents. Previously, when existing methods that include Ranking SVM were applied to document retrieval, none of the two factors was taken into consideration. We show it is possible to make modifications in conventional Ranking SVM, so it can be better used for document retrieval. Specifically, we modify the "Hinge Loss" function in Ranking SVM to deal with the problems described above. We employ two methods to conduct optimization on the loss function: gradient descent and quadratic programming. Experimental results show that our method, referred to as Ranking SVM for IR, can outperform the conventional Ranking SVM and other existing methods for document retrieval on two datasets.

This paper is concerned with 'intranet search'. By intranet search, we mean searching for information on an intranet within an organization. We have found that search needs on an intranet can be categorized into types, through an analysis of survey results and an analysis of search log data. The types include searching for definitions, persons, experts, and homepages. Traditional information retrieval only focuses on search of relevant documents, but not on search of special types of information. We propose a new approach to intranet search in which we search for information in each of the special types, in addition to the traditional relevance search. Information extraction technologies can play key roles in such kind of 'search by type' approach, because we must first extract from the documents the necessary information in each type. We have developed an intranet search system called 'Information Desk'. In the system, we try to address the most important types of search first - finding term definitions, homepages of groups or topics, employees' personal information and experts on topics. For each type of search, we use information extraction technologies to extract, fuse, and summarize information in advance. The system is in operation on the intranet of Microsoft and receives accesses from about 500 employees per month. Feedbacks from users and system logs show that users consider the approach useful and the system can really help people to find information. This paper describes the architecture, features, component technologies, and evaluation results of the system.