Large Language Models (LLMs) have shown strong capabilities in open-domain question answering (QA), but deploying them in real-world online systems introduces critical challenges. These include: (1) handling both simple and complex queries with appropriate levels of reasoning, (2) minimizing latency without compromising answer quality, and (3) maintaining answer consistency under evolving and noisy retrieval contexts. To address these challenges, we propose Retain-to-Refine (ℜ2ℜ), an adaptive agent-based QA framework designed for practical deployment. ℜ2ℜ integrates a Query Critic Agent (QCA) to assess query difficulty and route it accordingly: simple queries are answered directly using fast, prompt-based LLM calls, while complex queries are handled by a Memory Augmented Agent (MAA). MAA performs iterative reasoning guided by a unique long-short memory mechanism. Long-term memory retains and consolidates stable, core facts to ground the reasoning process, while short-term memory identifies transient information gaps to formulate highly focused subsequent queries. To ensure evidence quality, a Supervised Retrospection module validates and filters retrieved documents at each step. This agent-based design enables ℜ2ℜ to dynamically allocate computation based on question complexity, reducing unnecessary overhead while preserving high-quality answers when multi-step reasoning or external knowledge is required. Extensive evaluations across various settings and datasets demonstrate that the efficiency of R2R across diverse question types. In online settings, ℜ2ℜ delivers substantial gains in both response quality and efficiency, making it well-suited for large-scale industrial deployment in real-time QA services.

Retrieval-Augmented Generation (RAG) has demonstrated remarkable success in enhancing Large Language Models (LLMs) through external knowledge integration. However, its application has primarily focused on textual content, leaving the rich domain of multi-modal video knowledge predominantly unexplored. This paper introduces VideoRAG, a retrieval-augmented generation framework designed for processing extremely long-context videos. Our core innovation lies in a dual-channel architecture that integrates graph-based textual knowledge grounding for cross-video semantic relationships and multi-modal context encoding for visual feature preservation. This design empowers VideoRAG to process unlimited-length videos by constructing precise knowledge graphs spanning multiple videos. The framework maintains semantic dependencies through specialized multi-modal retrieval paradigms, enabling comprehensive understanding of extended video content. We evaluate VideoRAG on our proposed LongerVideos benchmark, comprising over 160 videos totaling 134+ hours across lecture, documentary, and entertainment categories. VideoRAG demonstrates substantial performance improvements compared to existing RAG alternatives and long video understanding methods. The source code for VideoRAG implementation, along with the benchmark dataset, are available at: https://github.com/HKUDS/VideoRAG.

Deep-research agents are capable of executing multi-step web exploration, targeted retrieval, and sophisticated question answering. Despite their powerful capabilities, deep-research agents face two critical bottlenecks: (1) the lack of large-scale, challenging datasets with real-world difficulty, and (2) the absence of accessible, open-source frameworks for data synthesis and agent training. To bridge these gaps, we first construct DeepResearch-9K, a large-scale challenging dataset specifically designed for deep-research scenarios built from open-source multi-hop question-answering (QA) datasets via a low-cost autonomous pipeline. Notably, it consists of (1) 9000 questions spanning three difficulty levels from L1 to L3 (2) high-quality search trajectories with reasoning chains from Tongyi-DeepResearch-30B-A3B, a state-of-the-art deep-research agent, and (3) verifiable answers. Furthermore, we develop an open-source training framework DeepResearch-R1 that supports (1) multi-turn web interactions, (2) different reinforcement learning (RL) approaches, and (3) different reward models such as rule-based outcome reward and LLM-as-Judge feedback. Finally, empirical results demonstrate that agents trained on DeepResearch-9K under our DeepResearch-R1 achieve state-of-the-art results on challenging deep-research benchmarks. We release the DeepResearch-9K dataset on https://huggingface.co/datasets/artillerywu/DeepResearch-9K and the code of DeepResearch-R1 on https://github.com/Applied-Machine-Learning-Lab/SIGIR2026_DeepResearch-R1.

Recommender systems (RecSys) are widely used across various modern digital platforms and have garnered significant attention. Traditional recommender systems usually focus only on fixed and simple recommendation scenarios, making it difficult to generalize to new and unseen recommendation tasks in an interactive paradigm. Recently, the advancement of large language models (LLMs) has revolutionized the foundational architecture of RecSys, driving their evolution into more intelligent and interactive personalized recommendation assistants. However, most existing studies rely on fixed task-specific prompt templates to generate recommendations and evaluate the performance of personalized assistants, which limits the comprehensive assessments of their capabilities. This is because commonly used datasets lack high-quality textual user queries that reflect real-world recommendation scenarios, making them unsuitable for evaluating LLM-based personalized recommendation assistants. To address this gap, we introduce RecBench+, a new dataset benchmark designed to assess LLMs' ability to handle intricate user recommendation needs in the era of LLMs. RecBench+ encompasses a diverse set of queries that span both hard conditions and soft preferences, with varying difficulty levels. We evaluated commonly used LLMs on RecBench+ and uncovered below findings: 1) LLMs demonstrate preliminary abilities to act as recommendation assistants, 2) LLMs are better at handling queries with explicitly stated conditions, while facing challenges with queries that require reasoning or contain misleading information. Our dataset has been released at https://github.com/jiani-huang/RecBenchPlus.

Generative inference with Large Language Models (LLMs) is the cornerstone of web-scale AI search, where queries are answered using vast, heterogeneous documents retrieved via Retrieval-Augmented Generation (RAG). This paradigm is critically bottlenecked by the cost of self-attention mechanism on long context. The sheer diversity of retrieved web content (multi-sourced, multi-lingual, multi-faceted) makes simple Key-Value (KV) cache optimizations with pre-fixed subsets ineffective, demanding a dynamic, content-aware approach. This challenge, however, introduces a classic chicken-and-egg problem: the model cannot foresee the necessary KV entries for attention without first inferring on the content, yet doing so on the full context is prohibitively expensive. This paper introduces P&F, a unified framework that resolves this dilemma through a core ''probe-and-fetch'' mechanism, which ingeniously integrates with speculative decoding -- an acceleration approach already adopted in web-scale AI search. The probe step repurposes the speculative draft model: while generating candidate tokens, it simultaneously probes the context to predict the most salient KV entries the large model will need for attention. The fetch step immediately acts on this prediction, asynchronously fetching these sparse entries from memory. This synergistic design piggybacks the probing step onto the drafting process, allowing the expensive gathering of a sparse KV cache to be fully masked. Crucially, this co-design breaks the sequential dependency bottleneck that cripples naive integrations of speculative decoding and prefetching due to synchronization issues. Extensive experiments show P&F significantly outperforms state-of-the-art methods in throughput and scalability, offering a practical, drop-in solution. Extensive offline evaluations across various settings and datasets demonstrate that P&F yields superior throughput and scalability compared to advanced baselines, while maintaining model quality across diverse models and scales. In online settings, P&F delivers substantial gains in throughput improvements while preserving response quality, making it well-suited for large-scale industrial deployment in real-time AI Search services.

Generative retrieval (GR) reformulates the Information Retrieval (IR) task as the generation of document identifiers (docIDs). Despite its promise, existing GR models exhibit poor generalization to newly added documents, often failing to generate the correct docIDs. While incremental training offers a straightforward remedy, it is computationally expensive, resource-intensive, and prone to catastrophic forgetting, thereby limiting the scalability and practicality of GR.

With the rise of AI in education, knowledge tracing (KT) has become important for modeling students’ knowledge from interaction data. However, existing methods still face three major challenges, including limited modeling of personalized exercise–concept relations, low robustness to noisy interactions, and inefficient training due to suboptimal data selection. To address these issues, we propose R 2 GCurL, a novel KT framework with two key designs. First, we recast KT as a graph classification problem and construct dynamic graphs from student responses, enabling the model to capture structural relations between exercises and concepts for more personalized KT. Second, we introduce a data-centric curriculum learning strategy based on dynamic graph entropy. Under our definition, pairwise dynamic graph entropy measures graph-transition continuity, where larger values indicate stronger structural similarity. Its sequence-level aggregation is used to derive a structure-aware difficulty signal for sample scheduling. On top of this, an RL-based scheduler further adapts batch selection based on model feedback and is especially beneficial under noisier and more unstable training regimes. Theoretical analysis shows that R 2 GCurL has lower computational complexity than existing graph-based KT models. Extensive experiments on five real-world datasets confirm its effectiveness, robustness, and generalizability, including as a plug-and-play enhancement for sequence-based KT models.

Retrieval-augmented Generation (RAG) integrates Large Language Models (LLMs) with retrievers to access external knowledge, improving the factuality of LLM generation in knowledge-grounded tasks. To optimize the RAG performance, most previous work independently fine-tunes the retriever to adapt to frozen LLMs or trains the LLMs to use documents retrieved by off-the-shelf retrievers, lacking end-to-end training supervision. Recent work addresses this limitation by jointly training these two components but relies on overly simplifying assumptions of document independence, which has been criticized for being far from real-world scenarios. Thus, effectively optimizing the overall RAG performance remains a critical challenge. We propose a Direct Retrieval-augmented Optimization ( DRO ) framework that enables end-to-end training of two key components: (i) a generative knowledge selection model and (ii) an LLM generator. DRO alternates between two phases: (i) document permutation estimation and (ii) re-weighted maximization, progressively improving RAG components through a variational approach. In the estimation step, we treat document permutation as a latent variable and directly estimate its distribution from the selection model by applying an importance sampling strategy. In the maximization step, we calibrate the optimization expectation using importance weights and jointly train the selection model and LLM generator. Our theoretical analysis reveals that DRO is analogous to policy-gradient methods in reinforcement learning. Extensive experiments conducted on five datasets illustrate that DRO outperforms the best baseline with 5–15% improvements in EM and F1. We also qualitatively analyze the stability, convergence, and variance of DRO. (Code is available on DRO GitHub ).

Retrieval-augmented generation (RAG) has demonstrated strong potential in enhancing large language models (LLMs) for complex, real-world question answering. However, existing RAG frameworks remain inadequate for temporal scenarios, primarily due to their inability to jointly model temporal constraints in both retrieval and reasoning. On the retrieval side, traditional approaches focus on semantic similarity, often returning outdated or temporally misaligned evidence. On the generation side, these systems frequently produce factually incorrect or hallucinated answers when confronted with incomplete or temporally inconsistent information. Motivated by the observed limitations, we propose ChronoReflect+, a temporal logic-aware RAG framework that incorporates hybrid temporal-aware retrieval and progressive multi-step reflection. Our method iteratively refines both retrieval and reasoning, identifying and bridging information gaps as context accumulates. Extensive experiments demonstrate that ChronoReflect+ significantly outperforms state-of-the-art RAG baselines-improving end-to-end accuracy by 15.2%-particularly on questions involving implicit time expressions and multi-hop reasoning.

The widely adopted Cranfield paradigm fails to adequately capture user satisfaction due to a weak relevance-satisfaction correlation. Additionally, constructing test collections incurs high relevance annotation costs. To address these two limitations, we aim to explore the use of large language models (LLMs) to generate multilevel usefulness labels. We propose CLUE, a user-centric evaluation method that explicitly incorporates users' search context and behavior information into LLMs. Inspired by ordinal regression, it employs a cascade structure tailored for multilevel usefulness judgments. Our study shows that using CLUE, LLMs can effectively assess usefulness when provided with search context and behavior, outperforming third-party labeling methods. We also conduct ablation studies to explore the impact of each component in CLUE. Finally, we utilize the usefulness labels generated by CLUE to predict user satisfaction. Real-world experiments reveal that incorporating CLUE's usefulness labels significantly enhances the performance of the satisfaction prediction model.

In the dynamic landscape of large-scale web search, Query-Driven Text Summarization (QDTS) aims to generate concise and informative summaries from textual documents based on a given query, which is essential for improving user engagement and facilitating rapid decision-making. Traditional extractive summarization models, based primarily on ranking candidate summary segments, have been the dominant approach in industrial applications. However, these approaches suffer from two key limitations: 1) The multi-stage pipeline often introduces cumulative information loss and architectural bottlenecks due to its weakest component; 2) Traditional models lack sufficient semantic understanding of both user queries and documents, particularly when dealing with complex search intents. In this study, we propose a novel framework to pioneer the application of generative models to address real-time QDTS in industrial web search. Our approach integrates large model distillation, supervised fine-tuning, direct preference optimization, and lookahead decoding to transform a lightweight model with only 0.1B parameters into a domain-specialized QDTS expert. Evaluated on multiple industry-relevant metrics, our model outperforms the production baseline and achieves a new state of the art. Furthermore, it demonstrates excellent deployment efficiency, requiring only 334 NVIDIA L20 GPUs to handle ~50,000 queries per second under 55~ms average latency per query.

Pre-trained language models (PLMs) have been successfully used to build high-performance ranking models for large-scale information retrieval systems. However, traditional PLM-based ranking approaches face two key challenges: (1) these models use both sparse and dense content (such as the query/title and content of documents) as inputs, which may require different attention allocations; and (2) traditional PLM-based ranking approaches have identified multiple objectives to gauge user satisfaction with ranking results, but integrating these objectives into the end-to-end training process and the subsequent feature updates and iterations usually involves significant computational resource overhead. In this paper, we propose a novel PLM-based ranking approach M2oE Rank, Multi-objective Mixture-of-Experts (MoE) enhanced Ranking. Specifically, M2oERank lever-ages a context-aware PLM-based hierarchical encoder to extract semantic relevance between the query and the document title and content, while allowing for separate dense and sparse attention for different inputs. With the extracted semantic relevance repre-sentations, multifacet user satisfaction features and task-specific annotations, M2oERank employs an MoE module to perform multi-objective pre-training of ranking models focused on user satisfaction. Finally, M2oERank uses a weight fusion module that fuses outputs from the above experts to predict ranking scores. Moreover, we present a three-stage offline training strategy and the online system workflow for deploying M2oERank at web-scale search. To demonstrate the effectiveness of our proposed approach, we conduct extensive offline and online evaluations using real-world web traffic from Baidu Search. The comparisons against numbers of advanced baselines confirmed the advantages of M2oERank in producing high-performance ranking models for web-scale search.

Benefiting from the strong reasoning capabilities, Large language models (LLMs) have demonstrated remarkable performance in recommender systems. Various efforts have been made to distill knowledge from LLMs to enhance collaborative models, employing techniques like contrastive learning for representation alignment. In this work, we prove that directly aligning the representations of LLMs and collaborative models is suboptimal for enhancing downstream recommendation tasks performance, based on the information theorem. Consequently, the challenge of effectively aligning semantic representations between collaborative models and LLMs remains unresolved. Inspired by this viewpoint, we propose a novel plug-and-play alignment framework for LLMs and collaborative models. Specifically, we first disentangle the latent representations of both LLMs and collaborative models into specific and shared components via projection layers and representation regularization. Subsequently, we perform both global and local structure alignment on the shared representations to facilitate knowledge transfer. Additionally, we theoretically prove that the specific and shared representations contain more pertinent and less irrelevant information, which can enhance the effectiveness of downstream recommendation tasks. Extensive experimental results on benchmark datasets demonstrate that our method is superior to existing state-of-the-art algorithms.

As modern learning-to-rank (LTR) systems rely on both textual and behavioral features, it is essential to extend pre-trained language models (PLMs) from text (queries and webpages) understanding to end-to-end ranking score prediction subject to multiple objectives, such as relevance, quality, authority, and recency. While textual inputs encompass a broader array of features than mere relevance and behavioral features are frequently skewed by user feedback with position bias, an integrated solution is required to jointly disentangle and fuse these heterogeneous features, ensuring robust and unbiased ranking predictions. In this work, we introduce RankExpert, a unified framework that holistically models heterogeneous ranking signals by integrating PLM-based semantic extraction with behavioral cues. RankExpert employs a lightweight PLM with hierarchical distillation for efficient query-document representation, a mixture-of-experts (MoE) architecture to disentangle and optimize objectives across relevance, quality, authority, and recency, and a click expert to mitigate position bias in user feedback within a dual-tower design. Then, RankExpert utilizes an adaptive weight fusion layer to dynamically integrate the outputs of these specialized experts dynamically, aligning ranking predictions with diverse user intents. Extensive offline evaluations on two large-scale real-world datasets reveal that RankExpert significantly outperforms strong competitor models and substantial gains in key performance indicators. Moreover, we deploy RankExpert at Baidu Search and conduct comprehensive online evaluations with real-world web traffic, demonstrating substantial improvements in user satisfaction metrics over the online legacy system and underscoring the practicality of the proposed approach.

The exponential growth of online content and increasingly diverse user needs have underscored the necessity for ranking models that go beyond traditional relevance assessments. Although several open-source benchmarks have significantly advanced academic research in Learning-to-Rank (LTR), these datasets predominantly focus on either text-based relevance or user behavior (click-through or dwell time) signals separately. This separation has inadvertently burdened academic progress by limiting the exploration of multifaceted, satisfaction-oriented ranking models. In contrast, industry research has begun to delve into integrated approaches that fuse prior (relevance, authority, recency, and quality) with posterior (user interaction such as clicks and dwell time) signals, thereby better capturing true user satisfaction. In this paper, we introduce FULTR-a large-scale, prior-posterior FUsion LTR dataset. FULTR comprises over 224M queries and 683M documents from Baidu Search, combining both: (1) a rich prior-attribute set with detailed textual relevance, authority, recency, and quality features, and (2) a comprehensive posterior-attribute set enriched by user click data, dwell time, and positional information. By unifying these dual perspectives, FULTR establishes a robust, reproducible benchmark for satisfaction-oriented ranking, enabling researchers to develop models that better capture real-world search behaviors and user satisfaction. In addition, we propose a strong LTR baseline that merges a satisfaction ranker that leverages pre-trained language models to integrate diverse satisfaction signals, with a behavior ranker that captures user interactions using a dual-tower approach. Their outputs are combined via a fusion layer, yielding significant performance gains in multiple evaluation metrics, as confirmed by extensive experiments and ablation studies. We are confident that our contribution not only democratizes access to industrial-grade fusion data for the research community but also paves the way for more effective and holistic LTR model design. FULTR is available to the research community at https://github.com/zhanghao731/FULTR.

The proliferation of complex non-factoid questions in modern information seeking (IS) systems exposes critical limitations in conventional Retrieval-Augmented Generation (RAG) approaches, particularly their static search strategies and the lack of systematic multi-source information integration capabilities. Facing these limitations, we present PASS (Proactive Agent-driven Search System), a novel multi-agent framework that operationalizes human-like proactive search strategies through five specialized agents: Revealer for intent analysis, Navigator for search planning, Seeker/Reader for adaptive retrieval, and Writer for response synthesis, systematically expanding the search space through iterative query refinement and multi-perspective knowledge integration. Crucially, our framework demonstrates remarkable adaptability to mid-sized LLMs, demonstrating its scalability in resource-constrained environments. To comprehensively assess the effectiveness of the proposed framework, we carry out extensive experiments on both mid-sized and proprietary large-scale LLMs, evaluating response quality for complex non-factoid questions using a newly introduced nugget-based assessment. Experimental results from offline nugget-based evaluation and online A/B Tests confirm substantial improvements in answer quality, advancing proactive information seeking methodologies and offering practical pathways for democratizing complex reasoning capabilities to resource-constrained environments.

While representation learning has been used to boost the performance of Learning-to-Rank (LTR) models through distilling key features for webpage ranking, the weak supervision signals extracted from users' sparse click-through data lead to inadequate representation of query-webpage pairs for ranking score prediction. Recent studies in generative LTR pre-training demonstrate the feasibility of incorporating reconstruction loss for enhanced ranking score prediction. However, LTR is afterall a regression task and it might be reasonable to find an alternate route that pre-trains LTR models with discriminative losses. Following the success of Electra in representation learning for natural language processing (NLP), this work proposes RankElectra that pre-trains the LTR model as a discriminator module inside a generative learning framework. Specifically, RankElectra first structures sparsely-annotated query-webpage pairs into a bipartite graph, with query and webpage feature vectors as node types and ranking scores as the connecting edges, and then leverages positive and negative extension strategies to densify the graph by link predictions. Later, this work proposes a novel Electra module that pre-trains the LTR model as a discriminator module for node reconstruction tasks, where node features of selected edges would be randomly masked and reconstructed by a generator, and the discriminator learns to classify whether the reconstructed features are the original or replaced as well as perform correct ranking. Finally, the pre-trained discriminator module, rather than the generator, would be fine-tuned on the labeled graph. We carried out extensive offline and online evaluations using the real-world web traffic of Baidu search engine. The results show that RankElectra could significantly boost the ranking performance of Baidu Search compared with numbers of competitor systems.

In book search, relevant book information should be returned in response to a query. Books contain complex, multi-faceted information such as metadata, outlines, and main text, where the outline provides hierarchical information between chapters and sections. Generative retrieval (GR) is a new retrieval paradigm that consolidates corpus information into a single model to generate identifiers of documents that are relevant to a given query. How can GR be applied to book search? Directly applying GR to book search is a challenge due to the unique characteristics of book search: The model needs to retain the complex, multi-faceted information of the book, which increases the demand for labeled data. Splitting book information and treating it as a collection of separate segments for learning might result in a loss of hierarchical information.

Heterogeneous Information Networks (HINs), which consist of various types of nodes and edges, have recently witnessed excellent performance in graph mining. However, most existing heterogeneous graph neural networks (HGNNs) fail to simultaneously handle the problems of missing attributes, inaccurate attributes and scarce node labels, which limits their expressiveness. In this paper, we propose a generative self-supervised model GraMI to address these issues simultaneously. Specifically, GraMI first initializes all the nodes in the graph with a low-dimensional representation matrix. After that, based on the variational graph autoencoder framework, GraMI learns both node-level and attribute-level embeddings in the encoder, which can provide fine-grained semantic information to construct node attributes. In the decoder, GraMI reconstructs both links and attributes. Instead of directly reconstructing raw features for attributed nodes, GraMI generates the initial low-dimensional representation matrix for all the nodes, based on which raw features of attributed nodes are further reconstructed. In this way, GraMI can not only complete informative features for non-attributed nodes, but rectify inaccurate ones for attributed nodes. Finally, we conduct extensive experiments to show the superiority of GraMI in tackling HINs with missing and inaccurate attributes. Our code and data can be found here: https://github.com/See-r/GraMI.

Generative retrieval (GR) has emerged as a promising paradigm in information retrieval (IR). However, most existing GR models are developed and evaluated using a static document collection, and their performance in dynamic corpora where document collections evolve continuously is rarely studied. In this paper, we first reproduce and systematically evaluate various representative GR approaches over dynamic corpora. Through extensive experiments, we reveal that existing GR models with text-based docids show superior generalization to unseen documents. We observe that the more fine-grained the docid design in the GR model, the better its performance over dynamic corpora, surpassing BM25 and even being comparable to dense retrieval methods. While GR models with numeric-based docids show high efficiency, their performance drops significantly over dynamic corpora. Furthermore, our experiments find that the underperformance of numeric-based docids is partly due to their excessive tendency toward the initial document set, which likely results from overfitting on the training set. We then conduct an in-depth analysis of the best-performing GR methods. We identify three critical advantages of text-based docids in dynamic corpora: (i) Semantic alignment with language models' pretrained knowledge (ii) Fine-grained docid design, and (iii) High lexical diversity. Building on these insights, we finally propose a novel multi-docid design that leverages both the efficiency of numeric-based docids and the effectiveness of text-based docids, achieving improved performance in dynamic corpus without requiring additional retraining. Our work offers empirical evidence for advancing GR methods over dynamic corpora and paves the way for developing more generalized yet efficient GR models in real-world search engines.

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.

Cross-Domain Recommendation (CDR) seeks to utilize knowledge from different domains to alleviate the problem of data sparsity in the target recommendation domain, and has been gaining more attention in recent years. Although there have been notable advances in this area, most current methods represent users and items in Euclidean space, which is not ideal for handling long-tail distributed data in recommendation systems. Additionally, adding data from other domains can worsen the long-tail characteristics of the entire dataset, making it harder to train CDR models effectively. Recent studies have shown that hyperbolic methods are particularly suitable for modeling long-tail distributions, which has led us to explore hyperbolic representations for users and items in CDR scenarios. However, due to the distinct characteristics of the different domains, applying hyperbolic representation learning to CDR tasks is quite challenging. In this paper, we introduce a new framework called Hyperbolic Contrastive Learning (HCTS), designed to capture the unique features of each domain while enabling efficient knowledge transfer between domains. We achieve this by embedding users and items from each domain separately and mapping them onto distinct hyperbolic manifolds with adjustable curvatures for prediction. To improve the representations of users and items in the target domain, we develop a hyperbolic contrastive learning module for knowledge transfer. Extensive experiments on real-world datasets demonstrate that hyperbolic manifolds are a promising alternative to Euclidean space for CDR tasks. The codes are available at https://github.com/EnkiXin/hcts.

Online learning to rank (OLTR), which directly optimizes the ranker with interactive user feedback, has gained considerable attention in both academia and industry. However, most current approaches suffer from the inefficiency of heuristic exploration strategies, which can seriously hurt users' experience. Furthermore, the existing OLTR solutions fail to learn from the cost-effective logged data, blocking their usage in the real industrial system. To handle the above issues, we in this paper introduce a new OLTR framework LT2R, namely Learning To online Learning to Rank. LT2R aims to study an efficient parameterized exploration strategy, by which a ranker could converge to the optimal ranking with as few exploration steps as possible. Specifically, we formulate the OLTR task as a typical Markov Decision Process and introduce an online reinforcement learning algorithm with a multi-round cumulative reward to guarantee fast convergence. Moreover, we contribute an offline learning algorithm for LT2R to exploit the knowledge from the historical searching logs, which can provide a fair warm-up model for its industrial deployment. Extensive experiments on both benchmark datasets and Baidu search engine have demonstrated its superiority over state-of-the-art methods.

Large language models (LLMs) have demonstrated remarkable capabilities across various research domains, including the field of Information Retrieval (IR). However, the responses generated by off-the-shelf LLMs tend to be generic, i.e., cannot capture the distinctiveness of each document with similar content. This limits the performance of LLMs in IR because finding and distinguishing relevant documents from substantial similar documents is a typical problem in many IR tasks. To address this issue, we propose an unsupervised alignment method, namely Reinforcement Learning from Contrastive Feedback (RLCF), empowering LLMs to generate both high-quality and context-specific responses. Our approach constructs unsupervised contrastive feedback signals based on similar document groups, and adopts a reward function, named group-wise reciprocal rank, to optimize LLMs. We conduct extensive experiments to evaluate the effectiveness of RLCF.

Large language models (LLMs) have demonstrated impressive capabilities in natural language processing. However, their internal mechanisms are still unclear and this lack of transparency poses unwanted risks for downstream applications. Therefore, understanding and explaining these models is crucial for elucidating their behaviors, limitations, and social impacts. In this article, we introduce a taxonomy of explainability techniques and provide a structured overview of methods for explaining Transformer-based language models. We categorize techniques based on the training paradigms of LLMs: traditional fine-tuning-based paradigm and prompting-based paradigm. For each paradigm, we summarize the goals and dominant approaches for generating local explanations of individual predictions and global explanations of overall model knowledge. We also discuss metrics for evaluating generated explanations and discuss how explanations can be leveraged to debug models and improve performance. Lastly, we examine key challenges and emerging opportunities for explanation techniques in the era of LLMs in comparison to conventional deep learning models.

Recommender systems have been widely adopted in various online personal e-commerce applications for improving user experience. A long-standing challenge in recommender systems is how to provide accurate recommendation to users in cold-start situations where only a few user-item interactions can be observed. Recently, meta learning methods provide a promising solution, and most of them follow a way of parameter initialization where predictions can be fast adapted via multiple gradient descent steps. While these meta-learning recommenders promote model performance, how to derive a fundamental paradigm that enables both flexible approximations of complex user interaction distributions and effective task adaptations of global knowledge still remains a critical yet under-explored problem. To this end, we present the Flow-based Adaptive Neural Process (FANP), a new probabilistic meta-learning model where estimating the preference of each user is governed by an underlying stochastic process. Following an encoder-decoder generative framework, FANP is an effective few-shot function estimator that directly maps limited user interactions to a predictive distribution without complicated gradient updates. Through introducing a conditional normalization flow-based encoder, FANP can get rid of the model bias on latent variables and thereby derive more flexible variational distributions. Meanwhile, we propose a task-adaptive mechanism capturing the relevance of different tasks for improving adaptation ability of global knowledge. The learned task-specific and task-relevant representations are simultaneously exploited to generate the decoder parameters via a novel modulation-augmented hypernetwork. FANP is evaluated on both scenario-specific and user-specific cold-start recommendations on various real-world datasets. Extensive experimental results and detailed model analyses demonstrate that our model yields superior performance compared with multiple state-of-the-art meta-learning recommenders.

User behavior data, such as ratings and clicks, has been widely used to build personalizing models for recommender systems. However, many unflattering factors (e.g., popularity, ranking position, users’ selection) significantly affect the performance of the learned recommendation model. Most existing work on unbiased recommendation addressed these biases from sample granularity (e.g., sample reweighting, data augmentation) or from the perspective of representation learning (e.g., bias-modeling). However, these methods are usually designed for a specific bias, lacking the universal capability to handle complex situations where multiple biases co-exist. Besides, rare work frees itself from laborious and sophisticated debiasing configurations (e.g., propensity scores, imputed values, or user behavior-generating process). Towards this research gap, in this article, we propose a universal G enerative framework for B ias D isentanglement termed as GBD , constantly generating calibration perturbations for the intermediate representations during training to keep them from being affected by the bias. Specifically, a bias-identifier that tries to retrieve the bias-related information from the representations is first introduced. Subsequently, the calibration perturbations are generated to significantly deteriorate the bias-identifier’s performance, making the bias gradually disentangled from the calibrated representations. Therefore, without relying on notorious debiasing configurations, a bias-agnostic model is obtained under the guidance of the bias identifier. We further present its universality by subsuming the representative biases and their mixture under the proposed framework. Finally, extensive experiments on the real-world, synthetic, and semi-synthetic datasets have demonstrated the superiority of the proposed approach against a wide range of recommendation debiasing methods. The code is available at https://github.com/Zhidan-Wang/GBD .

Passage retrieval is a fundamental task in many information systems, such as web search and question answering, where both efficiency and effectiveness are critical concerns. In recent years, neural retrievers based on pre-trained language models (PLM), such as dual-encoders, have achieved huge success. Yet, studies have found that the performance of dual-encoders are often limited due to the neglecting of the interaction information between queries and candidate passages. Therefore, various interaction paradigms have been proposed to improve the performance of vanilla dual-encoders. Particularly, recent state-of-the-art methods often introduce late-interaction during the model inference process. However, such late-interaction based methods usually bring extensive computation and storage cost on large corpus. Despite their effectiveness, the concern of efficiency and space footprint is still an important factor that limits the application of interaction-based neural retrieval models. To tackle this issue, we Incorporate Implicit Interaction into dual-encoders, and propose I3 retriever. In particular, our implicit interaction paradigm leverages generated pseudo-queries to simulate query-passage interaction, which jointly optimizes with query and passage encoders in an end-to-end manner. It can be fully pre-computed and cached, and its inference process only involves simple dot product operation of the query vector and passage vector, which makes it as efficient as the vanilla dual encoders. We conduct comprehensive experiments on MSMARCO and TREC2019 Deep Learning Datasets, demonstrating the I3 retriever's superiority in terms of both effectiveness and efficiency. Moreover, the proposed implicit interaction is compatible with special pre-training and knowledge distillation for passage retrieval, which brings a new state-of-the-art performance. The codes are available at https://github.com/Deriq-Qian-Dong/III-Retriever.

While learning to rank (LTR) is widely employed in web searches to prioritize pertinent webpages from the retrieved contents based on input queries, traditional LTR models stumble over two principal stumbling blocks leading to subpar performance: 1) the lack of well-annotated query-webpage pairs with ranking scores to cover search queries of various popularity, debilitating their coverage of search queries across the popularity spectrum, and 2) ill-trained models that are incapable of inducing generalized representations for LTR, culminating in overfitting. To tackle the above challenges, we proposed a Generative Semi-Supervised Pre-trained (GS2P) Learning to Rank model. Specifically, GS2P first generates pseudo-labels for the unlabeled samples using tree-based LTR models after a series of co-training procedures, then learns the representations of query-webpage pairs with self-attentive transformers via both discriminative (LTR) and generative (denoising autoencoding for reconstruction) losses. Finally, GS2P boosts the performance of LTR through incorporating Random Fourier Features to over-parameterize the models into “interpolating regime”, so as to enjoy the further descent of generalization errors with learned representations. We conduct extensive offline experiments on a publicly available dataset and a real-world dataset collected from a large-scale search engine. The results show that GS2P can achieve the best performance on both datasets, compared to baselines. We also deploy GS2P at a large-scale web search engine with realistic traffic, where we can still observe significant improvement in real-world applications. GS2P performs consistently in both online and offline experiments.

Both Transformer and Graph Neural Networks (GNNs) have been used in learning to rank (LTR), they however adhere to two distinct-yet-complementary problem formulations, i.e., ranking score regression based on query-webpage pairs and link prediction within query-webpage bipartite graphs, respectively. Though it is possible to pre-train GNNs or Transformers on source datasets and fine-tune them subject to sparsely annotated LTR datasets separately, the source-target distribution shifts across the pairs and bipartite graphs domains make it extremely difficult to integrate these diverse models into a single LTR framework at a web-scale. We introduce the novel MPGraf model, which utilizes a modular and capsule-based pre-training approach, aiming to incorporate regression capacities from Transformers and link prediction capabilities of GNNs cohesively. Specifically, rather than simply combining these two modules in stacking or parallelizing architectures, MPGraf proposes a three-step method to (1) construct query-webpage graphs from sparsely annotated query-webpage pairs, (2) pre-train the Transformer and GNN modules using source datasets from two domains, and (3) “surgically” fine-tune the integrated models with pretrained weights while handling the distribution shifts of two domains. Extensive experiments have been done to evaluate the performance of MPGraf using real-world datasets collected from large-scale search engines. The results show that MPGraf can outperform baseline algorithms on several major metrics. Further, we deploy and evaluate MPGraf atop a large-scale search engine with realistic web traffics via A/B tests, where we can still observe significant improvement. MPGraf performs consistently in both offline and online evaluations.

While Learning to Rank (LTR) models on top of transformers have been widely adopted to achieve decent performance, it is still challenging to train the model with sufficient data as only an extremely small number of query-webpage pairs could be annotated versus trillions of webpages available online and billions of web search queries everyday. In the meanwhile, industry research communities have released a number of open-source LTR datasets with well annotations but incorporating different designs of LTR features/labels (i.e., heterogeneous domains). In this work, inspired by the recent progress in pre-training transformers for performance advantages, we study the problem of pre-training LTR models using both labeled and unlabeled samples, especially we focus on the use of well-annotated samples in heterogeneous open-source LTR datasets to boost the performance of pre-training. Hereby, we propose S2phere-Semi-Supervised Pre-training with Heterogeneous LTR data strategies for LTR models using both unlabeled and labeled query-webpage pairs across heterogeneous LTR datasets. S2phere consists of a three-step approach: (1) Semi-supervised Feature Extraction Pre-training via Perturbed Contrastive Loss, (2) Cross-domain Ranker Pre-training over Heterogeneous LTR Datasets and (3) End-to-end LTR Fine-tuning via Modular Network Composition. Specifically, given an LTR model composed of a backbone (the feature extractor), a neck (the module to reason the orders) and a head (the predictor of ranking scores), S2phere uses unlabeled/labeled data from the search engine to pre-train the backbone in Step (1) via semi-supervised learning; then Step (2) incorporates multiple open-source heterogeneous LTR datasets to improve pre-training of the neck module as shared parameters of cross-domain learning; and finally, S2phere in Step (3) composes the backbone and neck with a randomly-initialized head into a whole LTR model and fine-tunes the model using search engine data with various learning strategies. Extensive experiments have been done with both offline experiments and online A/B Test on top of Baidu search engine. The comparisons against numbers of baseline algorithms confirmed the advantages of S2phere in producing high-performance LTR models for web-scale search.

Recently, a new paradigm called Differentiable Search Index (DSI) has been proposed for document retrieval, wherein a sequence-to-sequence model is learned to directly map queries to relevant document identifiers. The key idea behind DSI is to fully parameterize traditional ''index-retrieve'' pipelines within a single neural model, by encoding all documents in the corpus into the model parameters. In essence, DSI needs to resolve two major questions: (1) how to assign an identifier to each document, and (2) how to learn the associations between a document and its identifier. In this work, we propose a Semantic-Enhanced DSI model (SE-DSI) motivated by Learning Strategies in the area of Cognitive Psychology. Our approach advances original DSI in two ways: (1) For the document identifier, we take inspiration from Elaboration Strategies in human learning. Specifically, we assign each document an Elaborative Description based on the query generation technique, which is more meaningful than a string of integers in the original DSI; and (2) For the associations between a document and its identifier, we take inspiration from Rehearsal Strategies in human learning. Specifically, we select fine-grained semantic features from a document as Rehearsal Contents to improve document memorization. Both the offline and online experiments show improved retrieval performance over prevailing baselines.

Query understanding plays a key role in exploring users' search intents and facilitating users to locate their most desired information. However, it is inherently challenging since it needs to capture semantic information from short and ambiguous queries and often requires massive task-specific labeled data. In recent years, pre-trained language models (PLMs) have advanced various natural language processing tasks because they can extract general semantic information from large-scale corpora. However, directly applying them to query understanding is sub-optimal because existing strategies rarely consider to boost the search performance. On the other hand, search logs contain user clicks between queries and urls that provide rich users' search behavioral information on queries beyond their content. Therefore, in this paper, we aim to fill this gap by exploring search logs. In particular, we propose a novel graph-enhanced pre-training framework, GE-BERT, which leverages both query content and the query graph. The model is trained on a query graph where nodes are queries and two queries are connected if they lead to clicks on the same urls, to capture both semantic information and users' search behavioral information of queries. Extensive experiments on offline and online tasks have demonstrated the effectiveness of the proposed framework.

Whilelearning to rank(LTR) has been widely used in web search to prioritize most relevant webpages among the retrieved contents subject to the input queries, the traditional LTR models fail to deliver decent performance due to two main reasons: 1) the lack of well-annotated query-webpage pairs with ranking scores to cover search queries of various popularity, and 2) ill-trained models based on a limited number of training samples with poor generalization performance. To improve the performance of LTR models, tremendous efforts have been done from above two aspects, such as enlarging training sets with pseudo-labels of ranking scores by self-training, or refining the features used for LTR through feature extraction and dimension reduction. Though LTR performance has been marginally increased, we still believe these methods could be further improved in the newly-fashioned “interpolating regime”. Specifically, instead of lowering the number of features used for LTR models, our work proposes to transform original data with random Fourier feature, so as to over-parameterize the downstream LTR models (e.g., GBRank or LightGBM) with features in ultra-high dimensionality and achieve superb generalization performance. Furthermore, rather than self-training with pseudo-labels produced by the same LTR model in a “self-tuned” fashion, the proposed method incorporates the diversity of prediction results between the listwise and pointwise LTR models while co-training both models with a cyclic labeling-prediction pipeline in a “ping-pong” manner. We deploy the proposedCo-trained andOver-parameterizedLTRsystemCOLTRat Baidu search and evaluateCOLTRwith a large number of baseline methods. The results show thatCOLTRcould achieve$\Delta NDCG_{4}$= 3.64%$\sim$4.92%, compared to baselines, under various ratios of labeled samples. We also conduct a 7-day A/B Test using the realistic web traffics of Baidu Search, where we can still observe significant performance improvement around$\Delta NDCG_{4}$= 0.17%$\sim$0.92% in real-world applications.COLTRperforms consistently both in online and offline experiments.

Pre-trained language representation models (PLMs) such as BERT and Enhanced Representation through kNowledge IntEgration (ERNIE) have been integral to achieving recent improvements on various downstream tasks, including information retrieval. However, it is nontrivial to directly utilize these models for the large-scale web search due to the following challenging issues: (1) the prohibitively expensive computations of massive neural PLMs, especially for long texts in the web document, prohibit their deployments in the web search system that demands extremely low latency; (2) the discrepancy between existing task-agnostic pre-training objectives and the ad hoc retrieval scenarios that demand comprehensive relevance modeling is another main barrier for improving the online retrieval and ranking effectiveness; and (3) to create a significant impact on real-world applications, it also calls for practical solutions to seamlessly interweave the resultant PLM and other components into a cooperative system to serve web-scale data. Accordingly, we contribute a series of successfully applied techniques in tackling these exposed issues in this work when deploying the state-of-the-art Chinese pre-trained language model, i.e., ERNIE, in the online search engine system. We first present novel practices to perform expressive PLM-based semantic retrieval with a flexible poly-interaction scheme and cost-efficiently contextualize and rank web documents with a cheap yet powerful Pyramid-ERNIE architecture. We then endow innovative pre-training and fine-tuning paradigms to explicitly incentivize the query-document relevance modeling in PLM-based retrieval and ranking with the large-scale noisy and biased post-click behavioral data. We also introduce a series of effective strategies to seamlessly interwoven the designed PLM-based models with other conventional components into a cooperative system. Extensive offline and online experimental results show that our proposed techniques are crucial to achieving more effective search performance. We also provide a thorough analysis of our methodology and experimental results.

Extracting query-document relevance from the sparse, biased clickthrough log is among the most fundamental tasks in the web search system. Prior art mainly learns a relevance judgment model with semantic features of the query and document and ignores directly counterfactual relevance evaluation from the clicking log. Though the learned semantic matching models can provide relevance signals for tail queries as long as the semantic feature is available. However, such a paradigm lacks the capability to introspectively adjust the biased relevance estimation whenever it conflicts with massive implicit user feedback. The counterfactual evaluation methods, on the contrary, ensure unbiased relevance estimation with sufficient click information. However, they suffer from the sparse or even missing clicks caused by the long-tailed query distribution.

Passage re-ranking is to obtain a permutation over the candidate passage set from retrieval stage. Re-rankers have been boomed by Pre-trained Language Models (PLMs) due to their overwhelming advantages in natural language understanding. However, existing PLM based re-rankers may easily suffer from vocabulary mismatch and lack of domain specific knowledge. To alleviate these problems, explicit knowledge contained in knowledge graph is carefully introduced in our work. Specifically, we employ the existing knowledge graph which is incomplete and noisy, and first apply it in passage re-ranking task. To leverage a reliable knowledge, we propose a novel knowledge graph distillation method and obtain a knowledge meta graph as the bridge between query and passage. To align both kinds of embedding in the latent space, we employ PLM as text encoder and graph neural network over knowledge meta graph as knowledge encoder. Besides, a novel knowledge injector is designed for the dynamic interaction between text and knowledge encoder. Experimental results demonstrate the effectiveness of our method especially in queries requiring in-depth domain knowledge.

Deep pre-trained language models (e.g., BERT) lead to remarkable headway in many Natural Language Processing tasks. Their superior capacity in perceiving textual data is also witnessed in semantic matching tasks (e.g., question answering, web search). Particularly for matching a pair of query and text candidate, the current state-of-the-arts usually rely on the semantic representations produced by BERT, and compute relevance scores with various interaction (i.e., matching) methods. However, they may 1) miss fine-grained phrase-level interaction between the input query and candidate context or 2) lack a thoughtful consideration of both effectiveness and efficiency. Motivated by this, we propose \hyttInteractor, a BERT-based semantic matching model with a flexible contextualized interaction paradigm. It is capable of capturing fine-grained phrase-level information in the interaction, and thus is more effective to be applied for semantic matching tasks. Moreover, we further facilitate \hyttInteractor with a novel partial attention scheme, which significantly reduces the computational cost while maintaining the high effectiveness. We conduct comprehensive experimental evaluations on three datasets. The results show that \hyttInteractor achieves superior effectiveness and efficiency for semantic matching.

Retrieval is a crucial stage in web search that identifies a small set of query-relevant candidates from a billion-scale corpus. Discovering more semantically-related candidates in the retrieval stage is very promising to expose more high-quality results to the end users. However, it still remains non-trivial challenges of building and deploying effective retrieval models for semantic matching in real search engine. In this paper, we describe the retrieval system that we developed and deployed in Baidu Search. The system exploits the recent state-of-the-art Chinese pretrained language model, namely Enhanced Representation through kNowledge IntEgration (ERNIE), which facilitates the system with expressive semantic matching. In particular, we developed an ERNIE-based retrieval model, which is equipped with 1) expressive Transformer-based semantic encoders, and 2) a comprehensive multi-stage training paradigm. More importantly, we present a practical system workflow for deploying the model in web-scale retrieval. Eventually, the system is fully deployed into production, where rigorous offline and online experiments were conducted. The results show that the system can perform high-quality candidate retrieval, especially for those tail queries with uncommon demands. Overall, the new retrieval system facilitated by pretrained language model (i.e., ERNIE) can largely improve the usability and applicability of our search engine.

As the heart of a search engine, the ranking system plays a crucial role in satisfying users' information demands. More recently, neural rankers fine-tuned from pre-trained language models (PLMs) establish state-of-the-art ranking effectiveness. However, it is nontrivial to directly apply these PLM-based rankers to the large-scale web search system due to the following challenging issues: (1) the prohibitively expensive computations of massive neural PLMs, especially for long texts in the web document, prohibit their deployments in an online ranking system that demands extremely low latency; (2) the discrepancy between existing ranking-agnostic pre-training objectives and the ad-hoc retrieval scenarios that demand comprehensive relevance modeling is another main barrier for improving the online ranking system; (3) a real-world search engine typically involves a committee of ranking components, and thus the compatibility of the individually fine-tuned ranking model is critical for a cooperative ranking system. In this work, we contribute a series of successfully applied techniques in tackling these exposed issues when deploying the state-of-the-art Chinese pre-trained language model, i.e., ERNIE, in the online search engine system. We first articulate a novel practice to cost-efficiently summarize the web document and contextualize the resultant summary content with the query using a cheap yet powerful Pyramid-ERNIE architecture. Then we endow an innovative paradigm to finely exploit the large-scale noisy and biased post-click behavioral data for relevance-oriented pre-training. We also propose a human-anchored fine-tuning strategy tailored for the online ranking system, aiming to stabilize the ranking signals across various online components. Extensive offline and online experimental results show that the proposed techniques significantly boost the search engine's performance.

Post-click conversion, as a strong signal indicating the user preference, is salutary for building recommender systems. However, accurately estimating the post-click conversion rate (CVR) is challenging due to the selection bias, i.e., the observed clicked events usually happen on users' preferred items. Currently, most existing methods utilize counterfactual learning to debias recommender systems. Among them, the doubly robust (DR) estimator has achieved competitive performance by combining the error imputation based (EIB) estimator and the inverse propensity score (IPS) estimator in a doubly robust way. However, inaccurate error imputation may result in its higher variance than the IPS estimator. Worse still, existing methods typically use simple model-agnostic methods to estimate the imputation error, which are not sufficient to approximate the dynamically changing model-correlated target (i.e., the gradient direction of the prediction model). To solve these problems, we first derive the bias and variance of the DR estimator. Based on it, a more robust doubly robust (MRDR) estimator has been proposed to further reduce its variance while retaining its double robustness. Moreover, we propose a novel double learning approach for the MRDR estimator, which can convert the error imputation into the general CVR estimation. Besides, we empirically verify that the proposed learning scheme can further eliminate the high variance problem of the imputation learning. To evaluate its effectiveness, extensive experiments are conducted on a semi-synthetic dataset and two real-world datasets. The results demonstrate the superiority of the proposed approach over the state-of-the-art methods. The code is available at https://github.com/guosyjlu/MRDR-DL.

Recommender Systems have been playing essential roles in e-commerce portals. Existing recommendation algorithms usually learn the ranking scores of items by optimizing a single task (e.g. Click-through rate prediction) based on users' historical click sequences, but they generally pay few attention to simultaneously modeling users' multiple types of behaviors or jointly optimize multiple objectives (e.g. both Click-through rate and Conversion rate), which are both vital for e-commerce sites. In this paper, we argue that it is crucial to formulate users' different interests based on multiple types of behaviors and perform multi-task learning for significant improvement in multiple objectives simultaneously. We propose Deep Multifaceted Transformers (DMT), a novel framework that can model users' multiple types of behavior sequences simultaneously with multiple Transformers. It utilizes Multi-gate Mixture-of-Experts to optimize multiple objectives. Besides, it exploits unbiased learning to reduce the selection bias in the training data. Experiments on JD real production dataset demonstrate the effectiveness of DMT, which significantly outperforms state-of-art methods. DMT has been successfully deployed to serve the main traffic in the commercial Recommender System in JD.com. To facilitate future research, we release the codes and datasets at https://github.com/guyulongcs/CIKM2020_DMT.

Incorporating temporal information into recommender systems has recently attracted increasing attention from both the industrial and academic research communities. Existing methods mostly reduce the temporal information of behaviors to behavior sequences for subsequently RNN-based modeling. In such a simple manner, crucial time-related signals have been largely neglected. This paper aims to systematically investigate the effects of the temporal information in sequential recommendations. In particular, we firstly discover two elementary temporal patterns of user behaviors: "absolute time patterns'' and "relative time patterns'', where the former highlights user time-sensitive behaviors, e.g., people may frequently interact with specific products at certain time point, and the latter indicates how time interval influences the relationship between two actions. For seamlessly incorporating these information into a unified model, we devise a neural architecture that jointly learns those temporal patterns to model user dynamic preferences. Extensive experiments on real-world datasets demonstrate the superiority of our model, comparing with the state-of-the-arts.

Hierarchical user profiling that aims to model users' real-time interests in different granularity is an essential issue for personalized recommendations in E-commerce. On one hand, items (i.e. products) are usually organized hierarchically in categories, and correspondingly users' interests are naturally hierarchical on different granularity of items and categories. On the other hand, multiple granularity oriented recommendations become very popular in E-commerce sites, which require hierarchical user profiling in different granularity as well. In this paper, we propose HUP, a Hierarchical User Profiling framework to solve the hierarchical user profiling problem in E-commerce recommender systems. In HUP, we provide a Pyramid Recurrent Neural Networks, equipped with Behavior-LSTM to formulate users' hierarchical real-time interests at multiple scales. Furthermore, instead of simply utilizing users' item-level behaviors (e.g., ratings or clicks) in conventional methods, HUP harvests the sequential information of users' temporal finely-granular interactions (micro-behaviors, e.g., clicks on components of items like pictures or comments, browses with navigation of the search engines or recommendations) for modeling. Extensive experiments on two real-world E-commerce datasets demonstrate the significant performance gains of the HUP against state-of-the-art methods for the hierarchical user profiling and recommendation problems. We release the codes and datasets at https://github.com/guyulongcs/WSDM2020_HUP.

We introduce CitySearcher, a vertical search engine that searches for cities when queried for an interest. Generally in search engines, utilization of semantics between words is favorable for performance improvement. Even though ambiguous query words have multiple semantic meanings, search engines can return diversified results to satisfy different users' information needs. But for CitySearcher, mismatched semantic relationships can lead to extremely unsatisfactory results. For example, the city Sale would incorrectly rank high for the interest shopping because of semantic interpretations of the words. Thus in our system, the main challenge is to eliminate the mismatched semantic relationships resulting from the side effect of the semantic models. In the previous case, we aim to ignore the semantics of a city's name which is not indicative of the city's characteristics. In CitySearcher, we use word2vec, a very popular word embedding technique to estimate the semantics of the words and create the initial ranks of the cities. To reduce the effect of the mismatched semantic relationships, we generate a set of features for learning based on a novel clustering-based method. With the generated features, we then utilize learning to rank algorithms to rerank the cities for return. We use the English version of Wikivoyage dataset for evaluation of our system, where we sample a very small dataset for training. Experimental results demonstrate the performance gain of our system over various standard retrieval techniques.

A recommendation is called explainable if it not only predicts a numerical rating for an item, but also generates explanations for users' preferences. Most existing methods for explainable recommendation apply topic models to analyze user reviews to provide descriptions along with the recommendations they produce. So far, such methods have neglected user opinions and influences from social relations as a source of information for recommendations, even though these are known to improve the rating prediction.

In this study, we investigate diversified recommendation problem by supervised learning, seeking significant improvement in diversity while maintaining accuracy. In particular, we regard each user as a training instance, and heuristically choose a subset of accurate and diverse items as ground-truth for each user. We then represent each user or item as a vector resulted from the factorization of the user-item rating matrix. In our paper, we try to discover a factorization for matching the following supervised learning task. In doing this, we define two coupled optimization problems, parameterized matrix factorization and structural learning, to formulate our task. And we propose a diversified collaborative filtering algorithm (DCF) to solve the coupled problems. We also introduce a new pairwise accuracy metric and a normalized topic coverage diversity metric to measure the performance of accuracy and diversity respectively. Extensive experiments on benchmark datasets show the performance gains of DCF in comparison with the state-of-the-art algorithms.

Personalized search approaches tailor search results to users' current interests, so as to help improve the likelihood of a user finding relevant documents for their query. Previous work on personalized search focuses on using the content of the user's query and of the documents clicked to model the user's preference. In this paper we focus on a different type of signal: We investigate the use of behavioral information for the purpose of search personalization. That is, we consider clicks and dwell time for reranking an initially retrieved list of documents. In particular, we (i) investigate the impact of distributions of users and queries on document reranking; (ii) estimate the relevance of a document for a query at 2 levels, at the query‐level and at the word‐level, to alleviate the problem of sparseness; and (iii) perform an experimental evaluation both for users seen during the training period and for users not seen during training. For the latter, we explore the use of information from similar users who have been seen during the training period. We use the dwell time on clicked documents to estimate a document's relevance to a query, and perform Bayesian probabilistic matrix factorization to generate a relevance distribution of a document over queries. Our experiments show that: (i) for personalized ranking, behavioral information helps to improve retrieval effectiveness; and (ii) given a query, merging information inferred from behavior of a particular user and from behaviors of other users with a user‐dependent adaptive weight outperforms any combination with a fixed weight.

Query classification is an important part of exploring the characteristics of web queries. Existing studies are mainly based on Broder's classification scheme and classify user queries into navigational, informational, and transactional categories according to users' information needs. In this article, we present a novel classification scheme from the perspective of queries' temporal patterns. Queries' temporal patterns are inherent time series patterns of the search volumes of queries that reflect the evolution of the popularity of a query over time. By analyzing the temporal patterns of queries, search engines can more deeply understand the users' search intents and thus improve performance. Furthermore, we extract three groups of features based on the queries' search volume time series and use a support vector machine (SVM) to automatically detect the temporal patterns of user queries. Extensive experiments on the Million Query Track data sets of the Text REtrieval Conference (TREC) demonstrate the effectiveness of our approach.

Collaborative filtering (CF) is one of the most effective techniques in recommender systems, which can be either rating oriented or ranking oriented. Ranking-oriented CF algorithms demonstrated significant performance gains in terms of ranking accuracy, being able to estimate a precise preference ranking of items for each user rather than the absolute ratings (as rating-oriented CF algorithms do). Conventional memory-based ranking-oriented CF can be referred to as pairwise algorithms. They represent each user as a set of preferences on each pair of items for similarity calculations and predictions. In this study, we propose ListCF, a novel listwise CF paradigm that seeks improvement in both accuracy and efficiency in comparison with pairwise CF. In ListCF, each user is represented as a probability distribution of the permutations over rated items based on the Plackett-Luce model, and the similarity between users is measured based on the Kullback--Leibler divergence between their probability distributions over the set of commonly rated items. Given a target user and the most similar users, ListCF directly predicts a total order of items for each user based on similar users’ probability distributions over permutations of the items. Besides, we also reveal insightful connections among pointwise, pairwise, and listwise CF algorithms from the perspective of the matrix representations. In addition, to make our algorithm more scalable and adaptive, we present an incremental algorithm for ListCF, which allows incrementally updating the similarities between users when certain user submits a new rating or updates an existing rating. Extensive experiments on benchmark datasets in comparison with the state-of-the-art approaches demonstrate the promise of our approach.

Recently, ranking-oriented collaborative filtering (CF) algorithms have achieved great success in recommender systems. They obtained state-of-the-art performances by estimating a preference ranking of items for each user rather than estimating the absolute ratings on unrated items (as conventional rating-oriented CF algorithms do). In this paper, we propose a new ranking-oriented CF algorithm, called ListCF. Following the memory-based CF framework, ListCF directly predicts a total order of items for each user based on similar users' probability distributions over permutations of the items, and thus differs from previous ranking-oriented memory-based CF algorithms that focus on predicting the pairwise preferences between items. One important advantage of ListCF lies in its ability of reducing the computational complexity of the training and prediction procedures while achieving the same or better ranking performances as compared to previous ranking-oriented memory-based CF algorithms. Extensive experiments on three benchmark datasets against several state-of-the-art baselines demonstrate the effectiveness of our proposal.

Automatic image annotation plays a critical role in modern keyword‐based image retrieval systems. For this task, the nearest‐neighbor–based scheme works in two phases: first, it finds the most similar neighbors of a new image from the set of labeled images; then, it propagates the keywords associated with the neighbors to the new image. In this article, we propose a novel approach for image annotation, which simultaneously improves both phases of the nearest‐neighbor–based scheme. In the phase of neighbor search, different from existing work discovering the nearest neighbors with the predicted distance, we introduce a ranking‐oriented neighbor search mechanism (RNSM), where the ordering of labeled images is optimized directly without going through the intermediate step of distance prediction. In the phase of keyword propagation, different from existing work using simple heuristic rules to select the propagated keywords, we present a learning‐based keyword propagation strategy (LKPS), where a scoring function is learned to evaluate the relevance of keywords based on their multiple relations with the nearest neighbors. Extensive experiments on the Corel 5K data set and the MIR Flickr data set demonstrate the effectiveness of our approach.

Collaborative Filtering (CF) is one of the most successful algorithms in recommender systems. However, it suffers from data sparsity and scalability problems. Although many clustering techniques have been incorporated to alleviate these two problems, most of them fail to achieve further significant improvement in recommendation accuracy. First of all, most of them assume each user or item belongs to a single cluster. Since usually users can hold multiple interests and items may belong to multiple categories, it is more reasonable to assume that users and items can join multiple clusters (groups), where each cluster is a subset of like-minded users and items they prefer. Furthermore, most of the clustering-based CF models only utilize historical rating information in the clustering procedure but ignore other data resources in recommender systems such as the social connections of users and the correlations between items. In this article, we propose HMCoC, a Hybrid Multigroup CoClustering recommendation framework, which can cluster users and items into multiple groups simultaneously with different information resources. In our framework, we first integrate information of user--item rating records, user social networks, and item features extracted from the DBpedia knowledge base. We then use an optimization method to mine meaningful user--item groups with all the information. Finally, we apply the conventional CF method in each cluster to make predictions. By merging the predictions from each cluster, we generate the top-n recommendations to the target users for return. Extensive experimental results demonstrate the superior performance of our approach in top-n recommendation in terms of MAP, NDCG, and F1 compared with other clustering-based CF models.

We propose CCRank, the first parallel framework for learning to rank based on evolutionary algorithms (EA), aiming to significantly improve learning efficiency while maintaining accuracy. CCRank is based on cooperative coevolution (CC), a divide-and-conquer framework that has demonstrated high promise in function optimization for problems with large search space and complex structures. Moreover, CC naturally allows parallelization of sub-solutions to the decomposed sub-problems, which can substantially boost learning efficiency. With CCRank, we investigate parallel CC in the context of learning to rank. We implement CCRank with three EA-based learning to rank algorithms for demonstration. Extensive experiments on benchmark datasets in comparison with the state-of-the-art algorithms show the performance gains of CCRank in efficiency and accuracy.

Collaborative filtering (CF) is an effective technique addressing the information overload problem. CF approaches generally fall into two categories: rating based and ranking based. The former makes recommendations based on historical rating scores of items and the latter based on their rankings. Ranking-based CF has demonstrated advantages in recommendation accuracy, being able to capture the preference similarity between users even if their rating scores differ significantly. In this study, we propose VSRank, a novel framework that seeks accuracy improvement of ranking-based CF through adaptation of the vector space model. In VSRank, we consider each user as a document and his or her pairwise relative preferences as terms. We then use a novel degree-specialty weighting scheme resembling TF-IDF to weight the terms. Extensive experiments on benchmarks in comparison with the state-of-the-art approaches demonstrate the promise of our approach.

Automatic image annotation plays an important role in modern keyword-based image retrieval systems. Recently, many neighbor-based methods have been proposed and achieved good performance for image annotation. However, existing work mainly focused on exploring a distance metric learning algorithm to determine the neighbors of an image, and neglected the subsequent keyword propagation process. They usually used some simple heuristic propagation rules, and propagated each keyword independently without considering the inherent semantic coherence among keywords. In this paper, we propose a novel learning-based keyword propagation strategy and incorporate it into the neighbor-based method framework. In particular, we employ the structural SVM to learn a scoring function which can evaluate different candidate keyword sets for a test image. Moreover, we explicitly enforce the semantic coherence constraint for the propagated keywords in our approach. The annotation of the test image is propagated as a whole rather than separate keywords. Experiments on two benchmark data sets demonstrate the effectiveness of our approach for image annotation and ranked retrieval.

Most existing recommender systems can be classified into two categories: collaborative filtering and content-based filtering. Hybrid recommender systems combine the advantages of the two for improved recommendation performance. Traditional recommender systems are rating-based. However, predicting ratings is an intermediate step towards their ultimate goal of generating rankings or recommendation lists. Learning to rank is an established means of predicting rankings and has recently demonstrated high promise in improving quality of recommendations. In this paper, we propose LRHR, the first attempt that adapts learning to rank to hybrid recommender systems. LRHR first defines novel representations for both users and items so that they can be content-comparable. Then, LRHR identifies a set of novel meta-level features for learning purposes. Finally, LRHR adopts RankSVM, a pairwise learning to rank algorithm, to generate recommendation lists of items for users. Extensive experiments on benchmarks in comparison with the state-of-the-art algorithms demonstrate the performance gain of our approach.

In this paper, we introduce polygene-based evolution, a novel framework for evolutionary algorithms (EAs) that features distinctive operations in the evolution process. In traditional EAs, the primitive evolution unit is gene, where genes are independent components during evolution. In polygene-based evolutionary algorithms (PGEAs), the evolution unit is polygene, i.e., a set of co-regulated genes. Discovering and maintaining quality polygenes can play an effective role in evolving quality individuals. Polygenes generalize genes, and PGEAs generalize EAs. Implementing the PGEA framework involves three phases: polygene discovery, polygene planting, and polygene-compatible evolution. Extensive experiments on function optimization benchmarks in comparison with the conventional and state-of-the-art EAs demonstrate the potential of the approach in accuracy and efficiency improvement.

Collaborative filtering (CF) is an effective technique addressing the information overload problem. Recently ranking-based CF methods have shown advantages in recommendation accuracy, being able to capture the preference similarity between users even if their rating scores differ significantly. In this study, we seek accuracy improvement of ranking-based CF through adaptation of the vector space model, where we consider each user as a document and her pairwise relative preferences as terms. We then use a novel degree-specialty weighting scheme resembling TF-IDF to weight the terms. Then we use cosine similarity to select a neighborhood of users for the target user to make recommendations. Experiments on benchmarks in comparison with the state-of-the-art methods demonstrate the promise of our approach.

Importance weighted active learning (IWAL) introduces a weighting scheme to measure the importance of each instance for correcting the sampling bias of the probability distributions between training and test datasets. However, the weighting scheme of IWAL involves the distribution of the test data, which can be straightforwardly estimated in active learning by interactively querying users for labels of selected test instances, but difficult for conventional learning where there are no interactions with users, referred as passive learning. In this paper, we investigate the insufficient sampling bias problem, i.e., bias occurs only because of insufficient samples, but the sampling process is unbiased. In doing this, we present two assumptions on the sampling bias, based on which we propose a practical weighting scheme for the empirical loss function in conventional passive learning, and present IWPL, an importance weighted passive learning framework. Furthermore, we provide IWSVM, an importance weighted SVM for validation. Extensive experiments demonstrate significant advantages of IWSVM on benchmarks and synthetic datasets.

With an increasingly amount of information in web forums, quick comprehension of threads in web forums has become a challenging research problem. To handle this issue, this paper investigates the task of Web Forum Thread Summarization (WFTS), aiming to give a brief statement of each thread that involving multiple dynamic topics. When applied to the task of WFTS, traditional summarization methods are cramped by topic dependencies, topic drifting and text sparseness. Consequently, we explore an unsupervised topic propagation model in this paper, the Post Propagation Model (PPM), to burst through these problems by simultaneously modeling the semantics and the reply relationship existing in each thread. Each post in PPM is considered as a mixture of topics, and a product of Dirichlet distributions in previous posts is employed to model each topic dependencies during the asynchronous discussion. Based on this model, the task of WFTS is accomplished by extracting most significant sentences in a thread. The experimental results on two different forum data sets show that WFTS based on the PPM outperforms several state-of-the-art summarization methods in terms of ROUGE metrics.

Learning to rank represents a category of effective ranking methods for information retrieval. While the primary concern of existing research has been accuracy, learning efficiency is becoming an important issue due to the unprecedented availability of large-scale training data and the need for continuous update of ranking functions. In this paper, we investigate parallel learning to rank, targeting simultaneous improvement in accuracy and efficiency.

One fundamental issue of learning to rank is the choice of loss function to be optimized. Although the evaluation measures used in Information Retrieval (IR) are ideal ones, in many cases they can't be used directly because they do not satisfy the smooth property needed in conventional machine learning algorithms. In this paper a new method named RankCSA is proposed, which tries to use IR evaluation measure directly. It employs the clonal selection algorithm to learn an effective ranking function by combining various evidences in IR. Experimental results on the LETOR benchmarh datasets demonstrate that RankCSA outperforms the baseline methods in terms of [email protected], MAP and [email protected]

Nowadays ranking function discovery approaches using Evolutionary Computation (EC), especially Genetic Programming (GP), have become an important branch in the Learning to Rank for Information Retrieval (LR4IR) field. Inspired by the GP based learning to rank approaches, we provide a series of generalized definitions and a common framework for the application of EC in learning to rank research. Besides, according to the introduced framework, we propose RankIP, a ranking function discovery approach using Immune Programming (IP). Experimental results demonstrate that RankIP evidently outperforms the baselines.