Emojis are globally used non-verbal cues in digital communication, and extensive research has examined how large language models (LLMs) understand and utilize emojis across contexts. While usually associated with friendliness or playfulness, it is observed that emojis may trigger toxic content generation in LLMs. Motivated by such a observation, we aim to investigate: (1) whether emojis can clearly enhance the toxicity generation in LLMs and (2) how to interpret this phenomenon.* We begin with a comprehensive exploration of emoji-triggered LLM toxicity generation by automating the construction of prompts with emojis to subtly express toxic intent. Experiments across 5 mainstream languages on 7 famous LLMs along with jailbreak tasks demonstrate that prompts with emojis could easily induce toxicity generation. To understand this phenomenon, we conduct model-level interpretations spanning semantic cognition, sequence generation and tokenization, suggesting that emojis can act as a heterogeneous semantic channel to bypass the safety mechanisms. To pursue deeper insights, we further probe the pre-training corpus and uncover potential correlation between the emoji-related data polution with the toxicity generation behaviors.

Jiale Cheng, Yusen Liu, Xinyu Zhang, Yulin Fei, Wenyi Hong, Ruiliang Lyu, Weihan Wang, Zhe Su, Xiaotao Gu, Xiao Liu, Yushi Bai, Jie Tang, Hongning Wang, Minlie Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Andrew Zhuoer Feng, Cunxiang Wang, Yu Luo, Lin Fan, Irene Zhou, Zikang Wang, Xiaotao Gu, Jie Tang, Hongning Wang, Minlie Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Bosi Wen, Yilin Niu, Cunxiang Wang, Pei Ke, Xiaoying Ling, Ying Zhang, Aohan Zeng, Hongning Wang, Minlie Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Bosi Wen, Yilin Niu, Cunxiang Wang, Xiaoying Ling, Ying Zhang, Pei Ke, Hongning Wang, Minlie Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Junxiao Yang, Haoran Liu, Jinzhe Tu, Jiale Cheng, Zhexin Zhang, Shiyao Cui, Jiaqi Weng, Jialing Tao, Hui Xue, Hongning Wang, Han Qiu, Minlie Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Zhexin Zhang, Xian Qi Loye, Victor Shea-Jay Huang, Junxiao Yang, Qi Zhu, Shiyao Cui, Fei Mi, Lifeng Shang, Yingkang Wang, Hongning Wang, Minlie Huang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Erle Zhu, Dazhi Jiang, Yuan Wang, Xujun Li, Jiale Cheng, Yuxian Gu, Yilin Niu, Aohan Zeng, Jie Tang, Minlie Huang, Hongning Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Emotional support conversation (ESC) helps reduce people's psychological stress and provide emotional value through interactive dialogues. Due to the high cost of crowdsourcing a large ESC corpus, recent attempts use large language models for dialogue augmentation. However, existing approaches largely overlook the social dynamics inherent in ESC, leading to less effective simulations. In this paper, we introduce SocialSim, a novel framework that simulates ESC by integrating key aspects of social interactions: social disclosure and social awareness. On the seeker side, we facilitate social disclosure by constructing a comprehensive persona bank that captures diverse and authentic help-seeking scenarios. On the supporter side, we enhance social awareness by eliciting cognitive reasoning to generate logical and supportive responses. Building upon SocialSim, we construct SSConv, a large-scale synthetic ESC corpus of which quality can even surpass crowdsourced ESC data. We further train a chatbot on SSConv and demonstrate its state-of-the-art performance in both automatic and human evaluations. We believe SocialSim offers a scalable way to synthesize ESC, making emotional care more accessible and practical.

Character-based dialogue (aka role-playing) enables users to freely customize characters for interaction, which often relies on LLMs, raising the need to evaluate LLMs’ character customization capability. However, existing benchmarks fail to ensure a robust evaluation as they often only involve a single character category or evaluate limited dimensions. Moreover, the sparsity of character features in responses makes feature-focused generative evaluation both ineffective and inefficient. To address these issues, we propose CharacterBench, the largest bilingual generative benchmark, with 22,859 human-annotated samples covering 3,956 characters from 25 detailed character categories. We define 11 dimensions of 6 aspects, classified as sparse and dense dimensions based on whether character features evaluated by specific dimensions manifest in each response. We enable effective and efficient evaluation by crafting tailored queries for each dimension to induce characters’ responses related to specific dimensions. Further, we develop CharacterJudge model for cost-effective and stable evaluations. Experiments show its superiority over SOTA automatic judges (e.g., GPT-4) and our benchmark’s potential to optimize LLMs’ character customization.

Yida Lu, Jiale Cheng, Zhexin Zhang, Shiyao Cui, Cunxiang Wang, Xiaotao Gu, Yuxiao Dong, Jie Tang, Hongning Wang, Minlie Huang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Songjie Niu, Kaisen Yang, Rui Zhao, Yichao Liu, Zonglin Li, Hongning Wang, Wenguang Chen. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Jailbreaking attacks can effectively induce unsafe behaviors in Large Language Models (LLMs); however, the transferability of these attacks across different models remains limited. This study aims to understand and enhance the transferability of gradient-based jailbreaking methods, which are among the standard approaches for attacking white-box models. Through a detailed analysis of the optimization process, we introduce a novel conceptual framework to elucidate transferability and identify superfluous constraints—specifically, the response pattern constraint and the token tail constraint—as significant barriers to improved transferability. Removing these unnecessary constraints substantially enhances the transferability and controllability of gradient-based attacks. Evaluated on Llama-3-8B-Instruct as the source model, our method increases the overall Transfer Attack Success Rate (T-ASR) across a set of target models with varying safety levels from 18.4% to 50.3%, while also improving the stability and controllability of jailbreak behaviors on both source and target models.

Jinfeng Zhou, Yuxuan Chen, Yihan Shi, Xuanming Zhang, Leqi Lei, Yi Feng, Zexuan Xiong, Miao Yan, Xunzhi Wang, Yaru Cao, Jianing Yin, Shuai Wang, Quanyu Dai, Zhenhua Dong, Hongning Wang, Minlie Huang. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Existing retrieval-augmented code generation (RACG) methods typically use an external retrieval module to fetch semantically similar code snippets used for generating subsequent fragments.However, even for consecutive code fragments, the content often diverges due to logical progression, resulting in a content gap.This gap undermines the performance of current RACG methods, as external retrieval modules based on content matching fail to infer the specific information need of LLMs to generate the next code fragment.Therefore, we propose SelfRACG, a novel paradigm that enables large language models (LLMs) to Self-express their information needs to enhance RACG.Specifically, SelfRACG includes an information need expression module and a two-stage information need-guided training strategy, which encourages LLMs to express their information need.Extensive experiments demonstrate that SelfRACG can retrieve external knowledge that better aligns with the LLM's own information needs, resulting in superior generation performance compared to vanilla RACG.Moreover, both the training and deployment costs for retrieval in our framework are much lower than those of the strongest retrieval model. 1

Jinfeng Zhou, Yuxuan Chen, Jianing Yin, Yongkang Huang, Yihan Shi, Xikun Zhang, Libiao Peng, Rongsheng Zhang, Tangjie Lv, Zhipeng Hu, Hongning Wang, Minlie Huang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Video generation models have achieved remarkable progress in text-to-video tasks. These models are typically trained on text-video pairs with highly detailed and carefully crafted descriptions, while real-world user inputs during inference are often concise, vague, or poorly structured. This gap makes prompt optimization crucial for generating high-quality videos. Current methods often rely on large language models (LLMs) to refine prompts through in-context learning, but suffer from several limitations: they may distort user intent, omit critical details, or introduce safety risks. Moreover, they optimize prompts without considering the impact on the final video quality, which can lead to suboptimal results. To address these issues, we introduce VPO, a principled framework that optimizes prompts based on three core principles: harmlessness, accuracy, and helpfulness. The generated prompts faithfully preserve user intents and, more importantly, enhance the safety and quality of generated videos. To achieve this, VPO employs a two-stage optimization approach. First, we construct and refine a supervised fine-tuning (SFT) dataset based on principles of safety and alignment. Second, we introduce both text-level and video-level feedback to further optimize the SFT model with preference learning. Our extensive experiments demonstrate that VPO significantly improves safety, alignment, and video quality compared to baseline methods. Moreover, VPO shows strong generalization across video generation models. Furthermore, we demonstrate that VPO could outperform and be combined with RLHF methods on video generation models, underscoring the effectiveness of VPO in aligning video generation models. Our code and data are publicly available at https://github.com/thu-coai/VPO.

Instruction-following is a fundamental capability of language models, requiring the model to recognize even the most subtle requirements in the instructions and accurately reflect them in its output. Such an ability is well-suited for and often optimized by preference learning. However, existing methods often directly sample multiple independent responses from the model when creating preference pairs. Such practice can introduce content variations irrelevant to whether the instruction is precisely followed (e.g., different expressions about the same semantic), interfering with the goal of teaching models to recognize the key differences that lead to improved instruction following. In light of this, we introduce SPaR, a self-play framework integrating tree-search self-refinement to yield valid and comparable preference pairs free from distractions. By playing against itself, an LLM employs a tree-search strategy to refine its previous responses with respect to the instruction while minimizing unnecessary variations. Our experiments show that a LLaMA3-8B model, trained over three iterations guided by SPaR, surpasses GPT-4-Turbo on the IFEval benchmark without losing general capabilities. Furthermore, SPaR demonstrates promising scalability, greatly enhancing models like GLM-4-9B and LLaMA3-70B. We also identify how inference scaling in tree search would impact model performance. Our code and data are publicly available at https://github.com/thu-coai/SPaR.

This work investigates the selection of high-quality pre-training data from massive corpora to enhance LMs' capabilities for downstream usage. We formulate data selection as a generalized Optimal Control problem, which can be solved theoretically by Pontryagin's Maximum Principle (PMP), yielding a set of necessary conditions that characterize the relationship between optimal data selection and LM training dynamics. Based on these theoretical results, we introduce **P**MP-based **D**ata **S**election (**PDS**), a framework that approximates optimal data selection by solving the PMP conditions. In our experiments, we adopt PDS to select data from CommmonCrawl and show that the PDS-selected corpus accelerates the learning of LMs and constantly boosts their performance on a wide range of downstream tasks across various model sizes. Moreover, the benefits of PDS extend to ~400B models trained on ~10T tokens, as evidenced by the extrapolation of the test loss curves according to the Scaling Laws. PDS also improves data utilization when the pre-training data is limited, by reducing the data demand by 1.8 times, which helps mitigate the quick exhaustion of available web-crawled corpora. Our code, model, and data can be found at https://github.com/microsoft/LMOps/tree/main/data_selection.

Industrial recommendation systems (RS) rely on the multi-stage pipeline to balance effectiveness and efficiency when delivering items from a vast corpus to users. Existing RS benchmark datasets primarily focus on the exposure space, where novel RS algorithms are trained and evaluated. However, when these algorithms transition to real-world industrial RS, they face two critical challenges: (1) handling unexposed items—a significantly larger space than the exposed one, profoundly impacting their practical performance; and (2) overlooking the intricate interplay between multiple stages of the recommendation pipeline, resulting in suboptimal system performance. To bridge the gap between offline RS benchmarks and real-world online environments, we introduce RecFlow—an industrial full-flow recommendation dataset. Unlike existing datasets, RecFlow includes samples not only from the exposure space but also from unexposed items filtered at each stage of the RS funnel. RecFlow comprises 38 million interactions from 42,000 users across nearly 9 million items with additional 1.9 billion stage samples collected from 9.3 million online requests over 37 days and spanning 6 stages. Leveraging RecFlow, we conduct extensive experiments to demonstrate its potential in designing novel algorithms that enhance effectiveness by incorporating stage-specific samples. Some of these algorithms have already been deployed online at KuaiShou, consistently yielding significant gains. We propose RecFlow as the first comprehensive whole-pipeline benchmark dataset for the RS community, enabling research on algorithm design across the entire recommendation pipeline, including selection bias study, debiased algorithms, multi-stage consistency and optimality, multi-task recommendation, and user behavior modeling.

Despite the remarkable success of large language models (LLMs) on traditional natural language processing tasks, their planning ability remains a critical bottleneck in tackling complex multi-step reasoning tasks. Existing approaches mainly rely on prompting or task-specific fine-tuning, often suffering from weak robustness and cross-task generalization. To address the limitation, we introduce CodePlan, a scalable paradigm that empowers LLMs to generate and follow code-form plans---pseudocode that outlines high-level, structured reasoning processes. By leveraging the structured and versatile nature of code, CodePlan effectively captures the rich semantics and control flows inherent to sophisticated reasoning. Importantly, CodePlan allows the automatic extraction of code-form plans from massive, wide-ranging text corpora without the need for curated, task-specific datasets. This enables it to scale up efficiently and improve reasoning capabilities across diverse scenarios. To train CodePlan, we construct a large-scale dataset of 2M examples that integrate code-form plans with standard prompt-response pairs from existing corpora. With minimal computation overhead during both training and inference, CodePlan achieves a 25.1\% relative improvement compared with directly generating responses, averaged across 13 challenging multi-step reasoning benchmarks, spanning mathematical reasoning, symbolic reasoning, instruction-following, multi-hop QA, and decision-making tasks. Further analysis reveals CodePlan's increasing performance gains on more complex reasoning tasks, as well as significant data efficiency thanks to its generalization ability.

Pre-trained on extensive text and image corpora, current Multi-Modal Large Language Models (MLLM) have shown strong capabilities in general visual reasoning tasks. However, their performance is still lacking in physical domains that require understanding diagrams with complex physical structures and quantitative analysis based on multi-modal information. To address this, we develop a new framework, named **M**ulti-Modal Scientific Re**A**soning with **P**hysics Perception and **S**imulation (**MAPS**) based on an MLLM. MAPS decomposes expert-level multi-modal reasoning task into physical diagram understanding via a Physical Perception Model (PPM) and reasoning with physical knowledge via a simulator. The PPM module is obtained by fine-tuning a visual language model using carefully designed synthetic data with paired physical diagrams and corresponding simulation language descriptions. At the inference stage, MAPS integrates the simulation language description of the input diagram provided by PPM and results obtained through a Chain-of-Simulation process with MLLM to derive the underlying rationale and the final answer. Validated using our collected college-level circuit analysis problems, MAPS significantly improves reasoning accuracy of MLLM and outperforms all existing models. The results confirm MAPS offers a promising direction for enhancing multi-modal scientific reasoning ability of MLLMs. We will release our code, model and dataset used for our experiments upon publishing of this paper.

Large language models (LLMs) have been increasingly applied to various domains, which triggers increasing concerns about LLMs' safety on specialized domains, e.g. medicine. Despite prior explorations on general jailbreaking attacks, there are two challenges for applying existing attacks on testing the domain-specific safety of LLMs: (1) Lack of professional knowledge-driven attacks(2) Insufficient coverage of domain knowledge. To bridge this gap, we propose a new task, knowledge-to-jailbreak, which aims to generate jailbreaking attacks from domain knowledge, requiring both attack effectiveness and knowledge relevance. We collect a large-scale dataset with 12,974 knowledge-jailbreak pairs and fine-tune a large language model as jailbreak-generator, to produce domain knowledge-specific jailbreaks. Experiments on 13 domains and 8 target LLMs demonstrate the effectiveness of jailbreak-generator in generating jailbreaks that are both threatening to the target LLMs and relevant to the given knowledge. We also apply our method to an out-of-domain knowledge base, showing that jailbreak-generator can generate jailbreaks that are comparable in harmfulness to those crafted by human experts. Data and code are available at: https://github.com/THU-KEG/Knowledge-to-Jailbreak/.

Jailbreak attacks against multimodal large language Models (MLLMs) are a significant research focus. Current research predominantly focuses on maximizing attack success rate (ASR), often overlooking whether the generated responses actually fulfill the attacker's malicious intent. This oversight frequently leads to low-quality outputs that, while successful in bypassing safety filters, lack substantial harmful content. To address this gap, we propose JPS, Jailbreak MLLMs with collaborative visual Perturbation and textual Steering, which achieves jailbreaks via corporation of visual image and textually steering prompt. Specifically, JPS utilizes target-guided adversarial image perturbations for effective safety bypass, complemented by ''steering prompt'' optimized via a multi-agent system to specifically guide LLM responses fulfilling the attackers' intent. These visual and textual components undergo iterative co-optimization for enhanced performance. To evaluate the quality of attack outcomes, we propose the Malicious Intent Fulfillment Rate (MIFR) metric, assessed using a Reasoning-LLM-based evaluator. Our experiments show JPS sets a new state-of-the-art in both ASR and MIFR across various MLLMs and benchmarks, with analyses confirming its efficacy. Codes are available at https://github.com/thu-coai/JPS Warning: This paper contains potentially sensitive contents.

Toxicity detection in multimodal text-image content faces growing challenges, especially with multimodal implicit toxicity, where each modality appears benign on its own but conveys hazard when combined. Multimodal implicit toxicity appears not only as formal statements in social platforms but also prompts that can lead to toxic dialogs from Large Vision-Language Models (LVLMs). Despite the success in unimodal text or image moderation, toxicity detection for multimodal content, particularly the multimodal implicit toxicity, remains underexplored. To fill this gap, we comprehensively build a taxonomy for multimodal implicit toxicity (MMIT) and introduce an MMIT-dataset, comprising 2,100 multimodal statements and prompts across 7 risk categories (31 sub-categories) and 5 typical cross-modal correlation modes. To advance the detection of multimodal implicit toxicity, we build ShieldVLM, a model which identifies implicit toxicity in multimodal statements, prompts and dialogs via deliberative cross-modal reasoning. Experiments show that ShieldVLM outperforms existing strong baselines in detecting both implicit and explicit toxicity. The model and dataset will be publicly available to support future researches (Warning: This paper contains potentially sensitive contents). Warning: This paper contains potentially sensitive contents.

Retrieval-augmented generation (RAG) has emerged as a promising solution to enhance the reliability of large language models (LLMs) with external knowledge. Existing RAG methods share a common strategy for knowledge injection: they place the retrieved documents into the input context of the LLM, which we refer to as the in-context knowledge injection method. While this approach is simple and often effective, it has inherent limitations. Firstly, increasing the context length and number of relevant documents can lead to higher computational overhead and degraded performance, especially in complex reasoning tasks. More importantly, in-context knowledge injection operates primarily at the input level, but LLMs store their internal knowledge in their parameters. This gap fundamentally limits the capacity of in-context methods. To this end, we introduce Parametric RAG, a new RAG paradigm that integrates external knowledge directly into the feed-forward networks of an LLM through document parameterization. This approach not only reduces online computational costs by shortening the input context length, but also deepens the integration of external knowledge by enabling LLMs to utilize it in the same way as internal parametric knowledge. Experimental results demonstrate that Parametric RAG substantially enhances the effectiveness and efficiency of knowledge augmentation in LLMs. Also, it can be combined with in-context RAG methods to achieve even better performance. We have open-sourced all the code, data, and models in the following GitHub link: https://github.com/oneal2000/PRAG

Retrieval-Augmented Generation (RAG) systems have become a crucial tool to augment large language models (LLMs) with external knowledge for better task performance.However, existing traditional RAG methods inject knowledge directly into the context, resulting in several limitations.First, these methods highly rely on the in-context learning capability of LLMs, which often leads to excessively long contexts.This is inefficient due to the quadratic complexity of self-attention, leading to significant increase in inference time.Second, the extended context and the nature of self-attention can cause the LLMs to lose important information in the context, thereby degrading the original capabilities of LLMs.Third, the effectiveness of knowledge injection is perturbed by the permutation of knowledge within the extended context, reducing the robustness of existing RAG methods.To tackle the above problems, we propose DecoupledRAG, a method that decouples external knowledge from the context within the RAG framework.Specifically, we introduce a cross-attention based method that injects retrieved knowledge directly into the inference process of LLM on the fly, without modifying its parameters or the input context, so that the external knowledge can be utilized robustly in a permutation-independent manner.To the best of our knowledge, this is the first work that explore how to utilize cross-attention to inject knowledge with low training cost in decoder-only LLM era.By leveraging cross-attention operation, DecoupledRAG enables seamless knowledge aggregation without creating extended context.Experimental results demonstrate that our method could achieve

Genetic sequence identification from electrical characterization of single molecules has emerged as a promising alternative to traditional approaches. Since electrical data on single molecules is extremely noisy due to the limitations of even state-of-the-art approaches, achieving high detection rates is challenging, particularly when the task involves being able to distinguish a sequence from its single base-pair mismatches. To address this issue, we propose an architecture based on combining a convolutional neural network with an ensemble learning method, XGBoost. In addition, four different input feature representations are considered, 1D conductance probability distributions and 2D conductance versus distance probability distributions which can be viewed as images, with or without averaging over the experimental parameters. The with averaging case corresponds to feature matrices derived from mixed datasets. We find that 2D probability distributions are helpful with respect to classifier accuracy, but averaged conductance probability distributions are much more impactful and significantly enhance prediction accuracy. Our quantitative analysis of multiple sequences shows an impressive performance increase of approximately 10% for all sequences. While the basis of our analysis is conductance data of DNA strands and their single base-pair mismatches, our method is generally applicable to other single-molecule identification based on their conductance.

Meta-reinforcement learning (meta-RL) aims to quickly solve new RL tasks by leveraging knowledge from prior tasks. Previous studies often assume a single-mode homogeneous task distribution, ignoring possible structured heterogeneity among tasks. Such an oversight can hamper effective exploration and adaptation, especially with limited samples. In this work, we harness the structured heterogeneity among tasks via clustering to improve meta-RL, which facilitates knowledge sharing at the cluster level. To facilitate exploration, we also develop a dedicated cluster-level exploratory policy to discover task clusters via divide-and-conquer. The knowledge from the discovered clusters helps to narrow the search space of task-specific policy learning, leading to more sample-efficient policy adaptation. We evaluate the proposed method on environments with parametric clusters (e.g., rewards and state dynamics in the MuJoCo suite) and non-parametric clusters (e.g., control skills in the Meta-World suite). The results demonstrate strong advantages of our solution against a set of representative meta-RL methods.

Adversarial attacks against stochastic multi-armed bandit (MAB) algorithms have been extensively studied in the literature. In this work, we focus on reward poisoning attacks and find most existing attacks can be easily detected by our proposed detection method based on the test of homogeneity, due to their aggressive nature in reward manipulations. This motivates us to study the notion of stealthy attack against stochastic MABs and investigate the resulting attackability. Our analysis shows that against two popularly employed MAB algorithms, UCB1 and $\epsilon$-greedy, the success of a stealthy attack depends on the environmental conditions and the realized reward of the arm pulled in the first round. We also analyze the situation for general MAB algorithms equipped with our attack detection method and find that it is possible to have a stealthy attack that almost always succeeds. This brings new insights into the security risks of MAB algorithms.

Jiale Cheng, Xiao Liu, Kehan Zheng, Pei Ke, Hongning Wang, Yuxiao Dong, Jie Tang, Minlie Huang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Pei Ke, Bosi Wen, Andrew Feng, Xiao Liu, Xuanyu Lei, Jiale Cheng, Shengyuan Wang, Aohan Zeng, Yuxiao Dong, Hongning Wang, Jie Tang, Minlie Huang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Xiao Liu, Xuanyu Lei, Shengyuan Wang, Yue Huang, Andrew Feng, Bosi Wen, Jiale Cheng, Pei Ke, Yifan Xu, Weng Lam Tam, Xiaohan Zhang, Lichao Sun, Xiaotao Gu, Hongning Wang, Jing Zhang, Minlie Huang, Yuxiao Dong, Jie Tang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

When using language models (LMs) to solve complex problems, humans might struggle to understand the LM-generated solutions and repair the flawed ones.To assist humans in repairing them, we propose to automatically decompose complex solutions into multiple simpler pieces that correspond to specific subtasks.We introduce a novel objective for learning task decomposition, termed assistive value (AssistV), which measures the feasibility and speed for humans to repair the decomposed solution.We collect a dataset of human repair experiences on different decomposed solutions.Utilizing the collected data as in-context examples, we then learn to critique, refine, and rank decomposed solutions to improve AssistV.We validate our method under competitive programming problems: under 177 hours of human study, our method enables non-experts to solve 33.3% more problems, speeds them up by 3.3x, and empowers them to match unassisted experts.

While significant attention has been dedicated to exploiting weaknesses in LLMs through jailbreaking attacks, there remains a paucity of effort in defending against these attacks.We point out a pivotal factor contributing to the success of jailbreaks: the intrinsic conflict between the goals of being helpful and ensuring safety.Accordingly, we propose to integrate goal prioritization at both training and inference stages to counteract.Implementing goal prioritization during inference substantially diminishes the Attack Success Rate (ASR) of jailbreaking from 66.4% to 3.6% for ChatGPT.And integrating goal prioritization into model training reduces the ASR from 71.0% to 6.6% for Llama2-13B.Remarkably, even in scenarios where no jailbreaking samples are included during training, our approach slashes the ASR by half.Additionally, our findings reveal that while stronger LLMs face greater safety risks, they also possess a greater capacity to be steered towards defending against such attacks, both because of their stronger ability in instruction following.Our work thus contributes to the comprehension of jailbreaking attacks and defenses, and sheds light on the relationship between LLMs' capability and safety.Our code is available at https://github.com/thu-coai/ JailbreakDefense_GoalPriority.

The demand for collaborative and private bandit learning across multiple agents is surging due to the growing quantity of data generated from distributed systems. Federated bandit learning has emerged as a promising framework for private, efficient, and decentralized online learning. However, almost all previous works rely on strong assumptions of client homogeneity, i.e., all participating clients shall share the same bandit model; otherwise, they all would suffer linear regret. This greatly restricts the application of federated bandit learning in practice. In this work, we introduce a new approach for federated bandits for heterogeneous clients, which clusters clients for collaborative bandit learning under the federated learning setting. Our proposed algorithm achieves non-trivial sub-linear regret and communication cost for all clients, subject to the communication protocol under federated learning that at anytime only one model can be shared by the server.

Despite the remarkable advances in language modeling, current mainstream decoding methods still struggle to generate texts that align with human texts across different aspects. In particular, sampling-based methods produce less-repetitive texts which are often disjunctive in discourse, while search-based methods maintain topic coherence at the cost of increased repetition. Overall, these methods fall short in achieving holistic alignment across a broad range of aspects. In this work, we frame decoding from a language model as an optimization problem with the goal of strictly matching the expected performance with human texts measured by multiple metrics of desired aspects simultaneously. The resulting decoding distribution enjoys an analytical solution that scales the input language model distribution via a sequence-level energy function defined by these metrics. And most importantly, we prove that this induced distribution is guaranteed to improve the perplexity on human texts, which suggests a better approximation to the underlying distribution of human texts. To facilitate tractable sampling from this globally normalized distribution, we adopt the Sampling-Importance-Resampling technique. Experiments on various domains and model scales demonstrate the superiority of our method in metrics alignment with human texts and human evaluation over strong baselines.

To enhance the efficiency and practicality of federated bandit learning, recent advances have introduced incentives to motivate communication among clients, where a client participates only when the incentive offered by the server outweighs its participation cost. However, existing incentive mechanisms naively assume the clients are truthful: they all report their true cost and thus the higher cost one participating client claims, the more the server has to pay. Therefore, such mechanisms are vulnerable to strategic clients aiming to optimize their own utility by misreporting. To address this issue, we propose an incentive compatible (i.e., truthful) communication protocol, named Truth-FedBan, where the incentive for each participant is independent of its self-reported cost, and reporting the true cost is the only way to achieve the best utility. More importantly, Truth-FedBan still guarantees the sub-linear regret and communication cost without any overhead. In other words, the core conceptual contribution of this paper is, for the first time, demonstrating the possibility of simultaneously achieving incentive compatibility and nearly optimal regret in federated bandit learning. Extensive numerical studies further validate the effectiveness of our proposed solution.

The alignment of language models with human preferences is vital for their application in real-world tasks. The problem is formulated as optimizing the model's policy to maximize the expected reward that reflects human preferences with minimal deviation from the initial policy. While considered as a straightforward solution, reinforcement learning (RL) suffers from high variance in policy updates, which impedes efficient policy improvement. Recently, direct preference optimization (DPO) was proposed to directly optimize the policy from preference data. However, we show that DPO derived based on the optimal solution of the problem leads to a compromised mean-seeking approximation of the optimal solution in practice. In this paper, we propose efficient exact optimization (EXO) of the alignment objective. EXO is guaranteed to optimize in the same direction as RL algorithms asymptotically for arbitrary policy parametrization. This leads to the same mode-seeking solution, while enables efficient optimization by circumventing the complexities of RL. We also compare our method to DPO with both theoretical and empirical analyses, and further demonstrate the advantages of our method over existing approaches on realistic human preference data. Code is available at https://github.com/haozheji/exact-optimization.

The advent of generative AI (GenAI) technology produces a transformative impact on the content creation landscape, offering alternative approaches to produce diverse, good-quality content across media, thereby reshaping online ecosystems but also raising concerns about market over-saturation and the potential marginalization of human creativity. Our work introduces a competition model generalized from the Tullock contest to analyze the tension between human creators and GenAI. Our theory and simulations suggest that despite challenges, a stable equilibrium between human and AI-generated content is possible. Our work contributes to understanding the competitive dynamics in the content creation industry, offering insights into the future interplay between human creativity and technological advancements in GenAI.

states without platform intervention; 2. offline experiments employing our proposed intervention mechanisms on diverse datasets; and 3. results from a three-week online experiment conducted on Instagram Reels short-video recommendation platform.

The notable success of large language models (LLMs) has sparked an upsurge in building language agents to complete various complex tasks. We present AMOR, an agent framework based on open-source LLMs, which reasons with external knowledge bases and adapts to specific domains through human supervision to the reasoning process. AMOR builds reasoning logic over a finite state machine (FSM) that solves problems through autonomous executions and transitions over disentangled modules. This allows humans to provide direct feedback to the individual modules, and thus naturally forms process supervision. Based on this reasoning and feedback framework, we develop AMOR through two-stage fine-tuning: warm-up and adaptation. The former fine-tunes the LLM with examples automatically constructed from various public datasets, enabling AMOR to generalize across different knowledge environments, while the latter tailors AMOR to specific domains using process feedback. Extensive experiments across multiple domains demonstrate the advantage of AMOR to strong baselines, thanks to its FSM-based reasoning and process feedback mechanism. The code and data are publicly available at https://github.com/JianGuanTHU/AMOR.

Instruction following is one of the fundamental capabilities of large language models (LLMs). As the ability of LLMs is constantly improving, they have been increasingly applied to deal with complex human instructions in real-world scenarios. Therefore, how to evaluate the ability of complex instruction-following of LLMs has become a critical research problem. Existing benchmarks mainly focus on modeling different types of constraints in human instructions while neglecting the composition of different constraints, which is an indispensable constituent in complex instructions. To this end, we propose ComplexBench, a benchmark for comprehensively evaluating the ability of LLMs to follow complex instructions composed of multiple constraints. We propose a hierarchical taxonomy for complex instructions, including 4 constraint types, 19 constraint dimensions, and 4 composition types, and manually collect a high-quality dataset accordingly. To make the evaluation reliable, we augment LLM-based evaluators with rules to effectively verify whether generated texts can satisfy each constraint and composition. Furthermore, we obtain the final evaluation score based on the dependency structure determined by different composition types. ComplexBench identifies significant deficiencies in existing LLMs when dealing with complex instructions with multiple constraints composition.

On User-Generated Content (UGC) platforms, recommendation algorithms significantly impact creators' motivation to produce content as they compete for algorithmically allocated user traffic. This phenomenon subtly shapes the volume and diversity of the content pool, which is crucial for the platform's sustainability. In this work, we demonstrate, both theoretically and empirically, that a purely relevance-driven policy with low exploration strength boosts short-term user satisfaction but undermines the long-term richness of the content pool. In contrast, a more aggressive exploration policy may slightly compromise user satisfaction but promote higher content creation volume. Our findings reveal a fundamental trade-off between immediate user satisfaction and overall content production on UGC platforms. Building on this finding, we propose an efficient optimization method to identify the optimal exploration strength, balancing user and creator engagement. Our model can serve as a pre-deployment audit tool for recommendation algorithms on UGC platforms, helping to align their immediate objectives with sustainable, long-term goals.

Reinforcement Learning from Human Feedback (RLHF) has been pivotal in aligning Large Language Models with human values but often suffers from overoptimization due to its reliance on a proxy reward model. To mitigate this limitation, we first propose a lightweight uncertainty quantification method that assesses the reliability of the proxy reward using only the last layer embeddings of the reward model. Enabled by this efficient uncertainty quantification method, we formulate AdvPO, a distributionally robust optimization procedure to tackle the reward overoptimization problem in RLHF. Through extensive experiments on the Anthropic HH and TL;DR summarization datasets, we verify the effectiveness of AdvPO in mitigating the overoptimization problem, resulting in enhanced RLHF performance as evaluated through human-assisted evaluation.

This is the second workshop in the series which discusses the cutting-edge developments in research and applications of personalizing large language models (LLMs) and adapting them to the demands of diverse user populations and societal needs. The full-day workshop plan includes several keynotes and invited talks, a poster session and a panel discussion.

This workshop discusses the cutting-edge developments in research and applications of personalizing large language models (LLMs) and adapting them to the demands of diverse user populations and societal needs. The full-day workshop includes several keynotes and invited talks, a poster session and a panel discussion.

Repeated risk minimization is a popular choice in real-world recommender systems driving their recommendation algorithms to adapt to user preferences and trends. However, numerous studies have shown that it exacerbates retention disparities among user groups, resulting in polarization within the user population. Given the primary objective of improving long-term user engagement in most industrial recommender systems and the significant commercial benefits from a diverse user population, enforcing retention fairness across user population is therefore crucial. Nonetheless, this goal is highly challenging due to the unknown dynamics of user retention (e.g., when a user would abandon the system) and the simultaneous aim to maximize the experience of every user.

The Probability Ranking Principle (PRP) has been considered as the foundational standard in the design of information retrieval (IR) systems. The principle requires an IR module's returned list of results to be ranked with respect to the underlying user interests, so as to maximize the results' utility. Nevertheless, we point out that it is inappropriate to indiscriminately apply PRP through every stage of a contemporary IR system. Such systems contain multiple stages (e.g., retrieval, pre-ranking, ranking, and re-ranking stages, as examined in this paper). The selection bias inherent in the model of each stage significantly influences the results that are ultimately presented to users. To address this issue, we propose an improved ranking principle for multi-stage systems, namely the Generalized Probability Ranking Principle (GPRP), to emphasize both the selection bias in each stage of the system pipeline as well as the underlying interest of users. We realize GPRP via a unified algorithmic framework named Full Stage Learning to Rank. Our core idea is to first estimate the selection bias in the subsequent stages and then learn a ranking model that best complies with the downstream modules' selection bias so as to deliver its top ranked results to the final ranked list in the system's output. We performed extensive experiment evaluations of our developed Full Stage Learning to Rank solution, using both simulations and online A/B tests in one of the leading short-video recommendation platforms. The algorithm is proved to be effective in both retrieval and ranking stages. Since deployed, the algorithm has brought consistent and significant performance gain to the platform.

Nan Wang, Qifan Wang, Yi-Chia Wang, Maziar Sanjabi, Jingzhou Liu, Hamed Firooz, Hongning Wang, Shaoliang Nie. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

This study explores the impact of content creators’ competition on user welfare in recommendation platforms, as well as the long-term dynamics of relevance-driven recommendations. We establish a model of creator competition, under the setting where the platform uses a top-$K$ recommendation policy, user decisions are guided by the Random Utility model, and creators, in absence of explicit utility functions, employ arbitrary no-regret learning algorithms for strategy updates. We study the user welfare guarantee through the lens of Price of Anarchy and show that the fraction of user welfare loss due to creator competition is always upper bounded by a small constant depending on $K$ and randomness in user decisions; we also prove the tightness of this bound. Our result discloses an intrinsic merit of the relevance-driven recommendation policy, as long as users’ decisions involve randomness and the platform provides reasonably many alternatives to its users.

Recommendation systems aim to predict users' feedback on items not exposed to them yet. Confounding bias arises due to the presence of unmeasured variables (e.g., the socio-economic status of a user) that can affect both a user's exposure and feedback. Existing methods either (1) make untenable assumptions about these unmeasured variables or (2) directly infer latent confounders from users' exposure. However, they cannot guarantee the identification of counterfactual feedback, which can lead to biased predictions. In this work, we propose a novel method, i.e., identifiable deconfounder (iDCF), which leverages a set of proxy variables (e.g., observed user features) to resolve the aforementioned non-identification issue. The proposed iDCF is a general deconfounded recommendation framework that applies proximal causal inference to infer the unmeasured confounders and identify the counterfactual feedback with theoretical guarantees. Extensive experiments on various real-world and synthetic datasets verify the proposed method's effectiveness and robustness.

Conversational Recommender Systems (CRS) actively elicit user preferences to generate adaptive recommendations. Mainstream reinforcement learning-based CRS solutions heavily rely on handcrafted reward functions, which may not be aligned with user intent in CRS tasks. Therefore, the design of task-specific rewards is critical to facilitate CRS policy learning, which remains largely under-explored in the literature. In this work, we propose a novel approach to address this challenge by learning intrinsic rewards from interactions with users. Specifically, we formulate intrinsic reward learning as a multi-objective bi-level optimization problem. The inner level optimizes the CRS policy augmented by the learned intrinsic rewards, while the outer level drives the intrinsic rewards to optimize two CRS-specific objectives: maximizing the success rate and minimizing the number of turns to reach a successful recommendation}in conversations. To evaluate the effectiveness of our approach, we conduct extensive experiments on three public CRS benchmarks. The results show that our algorithm significantly improves CRS performance by exploiting informative learned intrinsic rewards.

Most existing works on federated bandits take it for granted that all clients are altruistic about sharing their data with the server for the collective good whenever needed. Despite their compelling theoretical guarantee on performance and communication efficiency, this assumption is overly idealistic and oftentimes violated in practice, especially when the algorithm is operated over self-interested clients, who are reluctant to share data without explicit benefits. Negligence of such self-interested behaviors can significantly affect the learning efficiency and even the practical operability of federated bandit learning. In light of this, we aim to spark new insights into this under-explored research area by formally introducing an incentivized communication problem for federated bandits, where the server shall motivate clients to share data by providing incentives. Without loss of generality, we instantiate this bandit problem with the contextual linear setting and propose the first incentivized communication protocol, namely, Inc-FedUCB, that achieves near-optimal regret with provable communication and incentive cost guarantees. Extensive empirical experiments on both synthetic and real-world datasets further validate the effectiveness of the proposed method across various environments.

The past decade has witnessed the flourishing of a new profession as media content creators, who rely on revenue streams from online content recommendation platforms. The reward mechanism employed by these platforms creates a competitive environment among creators which affects their production choices and, consequently, content distribution and system welfare. It is thus crucial to design the platform's reward mechanism in order to steer the creators' competition towards a desirable welfare outcome in the long run. This work makes two major contributions in this regard: first, we uncover a fundamental limit about a class of widely adopted mechanisms, coined \emph{Merit-based Monotone Mechanisms}, by showing that they inevitably lead to a constant fraction loss of the optimal welfare. To circumvent this limitation, we introduce \emph{Backward Rewarding Mechanisms} (BRMs) and show that the competition game resultant from BRMs possesses a potential game structure. BRMs thus naturally induce strategic creators' collective behaviors towards optimizing the potential function, which can be designed to match any given welfare metric. In addition, the class of BRM can be parameterized so that it allows the platform to directly optimize welfare within the feasible mechanism space even when the welfare metric is not explicitly defined.

Off-policy learning, referring to the procedure of policy optimization with access only to logged feedback data, has shown importance in various important real-world applications, such as search engines and recommender systems. While the ground-truth logging policy is usually unknown, previous work simply takes its estimated value for the off-policy learning, ignoring the negative impact from both high bias and high variance resulted from such an estimator. And these impact is often magnified on samples with small and inaccurately estimated logging probabilities. The contribution of this work is to explicitly model the uncertainty in the estimated logging policy, and propose an Uncertainty-aware Inverse Propensity Score estimator (UIPS) for improved off-policy learning, with a theoretical convergence guarantee. Experiment results on the synthetic and real-world recommendation datasets demonstrate that UIPS significantly improves the quality of the discovered policy, when compared against an extensive list of state-of-the-art baselines.

Contextual bandit algorithms have been popularly used to address interactive recommendation, where the users are assumed to be cooperative to explore all recommendations from a system. In this paper, we relax this strong assumption and study the problem of incentivized exploration with myopic users, where the users are only interested in recommendations with their currently highest estimated reward. As a result, in order to obtain long-term optimality, the system needs to offer compensation to incentivize the users to take the exploratory recommendations. We consider a new and practically motivated setting where the context features employed by the user are more informative than those used by the system: for example, features based on users’ private information are not accessible by the system. We develop an effective solution for incentivized exploration under such an information gap, and prove that the method achieves a sublinear rate in both regret and compensation. We theoretically and empirically analyze the added compensation due to the information gap, compared with the case where the system has access to the same context features as the user does, i.e., without information gap. Moreover, we also provide a compensation lower bound of this problem.

Advertisements (ads) are an innate part of search engine business models. To promote sales, advertisers are willing to pay search engines to promote their content to a prominent position in the search result page (SERP). This raises concerns about the search engine manipulation effect (SEME): the opinions of users can be influenced by the way search results are presented

In smart computing, the labels of training samples for a specific task are not always abundant. However, the labels of samples in a relevant but different dataset are available. As a result, researchers have relied on unsupervised domain adaptation to leverage the labels in a dataset (the source domain) to perform better classification in a different, unlabeled dataset (target domain). Existing non-generative adversarial solutions for UDA aim at achieving domain confusion through adversarial training. The ideal scenario is that perfect domain confusion is achieved, but this is not guaranteed to be true. To further enforce domain confusion on top of the adversarial training, we propose a novel UDA algorithm, E-ADDA, which uses both a novel variation of the Mahalanobis distance loss and an out-of-distribution detection subroutine. The Mahalanobis distance loss minimizes the distribution-wise distance between the encoded target samples and the distribution of the source domain, thus enforcing additional domain confusion on top of adversarial training. Then, the OOD subroutine further eliminates samples on which the domain confusion is unsuccessful. We have performed extensive and comprehensive evaluations of E-ADDA in the acoustic and computer vision modalities. In the acoustic modality, E-ADDA outperforms several state-of-the-art UDA algorithms by up to 29.8%, measured in the f1 score. In the computer vision modality, the evaluation results suggest that we achieve new state-of-the-art performance on popular UDA benchmarks such as Office-31 and Office-Home, outperforming the second best-performing algorithms by up to 17.9%.

Conversational recommender systems (CRS) explicitly solicit users' preferences for improved recommendations on the fly. Most existing CRS solutions count on a single policy trained by reinforcement learning for a population of users. However, for users new to the system, such a global policy becomes ineffective to satisfy them, i.e., the cold-start challenge. In this paper, we study CRS policy learning for cold-start users via meta reinforcement learning. We propose to learn a meta policy and adapt it to new users with only a few trials of conversational recommendations. To facilitate fast policy adaptation, we design three synergetic components. Firstly, we design a meta-exploration policy dedicated to identifying user preferences via a few exploratory conversations, which accelerates personalized policy adaptation from the meta policy. Secondly, we adapt the item recommendation module for each user to maximize the recommendation quality based on the collected conversation states during conversations. Thirdly, we propose a Transformer-based state encoder as the backbone to connect the previous two components. It provides comprehensive state representations by modeling complicated relations between positive and negative feedback during the conversation. Extensive experiments on three datasets demonstrate the advantage of our solution in serving new users, compared with a rich set of state-of-the-art CRS solutions.

Accuracy and diversity have long been considered to be two conflicting goals for recommendations. We point out, however, that as the diversity is typically measured by certain pre-selected item attributes, e.g., category as the most popularly employed one, improved diversity can be achieved without sacrificing recommendation accuracy, as long as the diversification respects the user's preference about the pre-selected attributes. This calls for a fine-grained understanding of a user's preferences over items, where one needs to recognize the user's choice is driven by the quality of the item itself, or the pre-selected attributes of the item.

We propose a new problem setting to study the sequential interactions between a recommender system and a user. Instead of assuming the user is omniscient, static, and explicit, as the classical practice does, we sketch a more realistic user behavior model, under which the user: 1) rejects recommendations if they are clearly worse than others; 2) updates her utility estimation based on rewards from her accepted recommendations; 3) withholds realized rewards from the system. We formulate the interactions between the system and such an explorative user in a K-armed bandit framework and study the problem of learning the optimal recommendation on the system side. We show that efficient system learning is still possible but is more difficult. In particular, the system can identify the best arm with probability at least 1-delta within O(1/delta) interactions, and we prove this is tight. Our finding contrasts the result for the problem of best arm identification with fixed confidence, in which the best arm can be identified with probability 1-delta within O(log(1/delta)) interactions. This gap illustrates the inevitable cost the system has to pay when it learns from an explorative user's revealed preferences on its recommendations rather than from the realized rewards.

Linear contextual bandit is a popular online learning problem. It has been mostly studied in centralized learning settings. With the surging demand of large-scale decentralized model learning, e.g., federated learning, how to retain regret minimization while reducing communication cost becomes an open challenge. In this paper, we study linear contextual bandit in a federated learning setting. We propose a general framework with asynchronous model update and communication for a collection of homogeneous clients and heterogeneous clients, respectively. Rigorous theoretical analysis is provided about the regret and communication cost under this distributed learning framework; and extensive empirical evaluations demonstrate the effectiveness of our solution.

Conversational recommender systems (CRS) dynamically obtain the users' preferences via multi-turn questions and answers. The existing CRS solutions are widely dominated by deep reinforcement learning algorithms. However, deep reinforcement learning methods are often criticized for lacking interpretability and requiring a large amount of training data to perform.

We study adversarial attacks on linear stochastic bandits: by manipulating the rewards, an adversary aims to control the behaviour of the bandit algorithm. Perhaps surprisingly, we first show that some attack goals can never be achieved. This is in a sharp contrast to context-free stochastic bandits, and is intrinsically due to the correlation among arms in linear stochastic bandits. Motivated by this finding, this paper studies the attackability of a $k$-armed linear bandit environment. We first provide a complete necessity and sufficiency characterization of attackability based on the geometry of the arms’ context vectors. We then propose a two-stage attack method against LinUCB and Robust Phase Elimination. The method first asserts whether the given environment is attackable; and if yes, it poisons the rewards to force the algorithm to pull a target arm linear times using only a sublinear cost. Numerical experiments further validate the effectiveness and cost-efficiency of the proposed attack method.

In real-world recommendation problems, especially those with a formidably large item space, users have to gradually learn to estimate the utility of any fresh recommendations from their experience about previously consumed items. This in turn affects their interaction dynamics with the system and can invalidate previous algorithms built on the omniscient user assumption. In this paper, we formalize a model to capture such ”learning users” and design an efficient system-side learning solution, coined Noise-Robust Active Ellipsoid Search (RAES), to confront the challenges brought by the non-stationary feedback from such a learning user. Interestingly, we prove that the regret of RAES deteriorates gracefully as the convergence rate of user learning becomes worse, until reaching linear regret when the user’s learning fails to converge. Experiments on synthetic datasets demonstrate the strength of RAES for such a contemporaneous system-user learning problem. Our study provides a novel perspective on modeling the feedback loop in recommendation problems.

Most real-world optimization problems have multiple objectives. A system designer needs to find a policy that trades off these objectives to reach a desired operating point. This problem has been studied extensively in the setting of known objective functions. However, we consider a more practical but challenging setting of unknown objective functions. In industry, optimization under this setting is mostly approached with online A/B testing, which is often costly and inefficient. As an alternative, we propose Interactive Multi-Objective Off-policy Optimization (IMO^3). The key idea of IMO^3 is to interact with a system designer using policies evaluated in an off-policy fashion to uncover which policy maximizes her unknown utility function. We theoretically show that IMO^3 identifies a near-optimal policy with high probability, depending on the amount of designer's feedback and training data for off-policy estimation. We demonstrate its effectiveness empirically on several multi-objective optimization problems.

Graph Convolutional Networks (GCNs) have fueled a surge of research interest due to their encouraging performance on graph learning tasks, but they are also shown vulnerability to adversarial attacks. In this paper, an effective graph structural attack is investigated to disrupt graph spectral filters in the Fourier domain, which are the theoretical foundation of GCNs. We define the notion of spectral distance based on the eigenvalues of graph Laplacian to measure the disruption of spectral filters. We realize the attack by maximizing the spectral distance and propose an efficient approximation to reduce the time complexity brought by eigen-decomposition. The experiments demonstrate the remarkable effectiveness of the proposed attack in both black-box and white-box settings for both test-time evasion attacks and training-time poisoning attacks. Our qualitative analysis suggests the connection between the imposed spectral changes in the Fourier domain and the attack behavior in the spatial domain, which provides empirical evidence that maximizing spectral distance is an effective way to change the graph structural property and thus disturb the frequency components for graph filters to affect the learning of GCNs.

Contextual bandit algorithms have been recently studied under the federated learning setting to satisfy the demand of keeping data decentralized and pushing the learning of bandit models to the client side. But limited by the required communication efficiency, existing solutions are restricted to linear models to exploit their closed-form solutions for parameter estimation. Such a restricted model choice greatly hampers these algorithms' practical utility. In this paper, we take the first step to addressing this challenge by studying generalized linear bandit models under the federated learning setting. We propose a communication-efficient solution framework that employs online regression for local update and offline regression for global update. We rigorously proved, though the setting is more general and challenging, our algorithm can attain sub-linear rate in both regret and communication cost, which is also validated by our extensive empirical evaluations.

We tackle the communication efficiency challenge of learning kernelized contextual bandits in a distributed setting. Despite the recent advances in communication-efficient distributed bandit learning, existing solutions are restricted to simple models like multi-armed bandits and linear bandits, which hamper their practical utility. In this paper, instead of assuming the existence of a linear reward mapping from the features to the expected rewards, we consider non-linear reward mappings, by letting agents collaboratively search in a reproducing kernel Hilbert space (RKHS). This introduces significant challenges in communication efficiency as distributed kernel learning requires the transfer of raw data, leading to a communication cost that grows linearly w.r.t. time horizon $T$. We addresses this issue by equipping all agents to communicate via a common Nystr\"{o}m embedding that gets updated adaptively as more data points are collected. We rigorously proved that our algorithm can attain sub-linear rate in both regret and communication cost.

Bandit algorithms have become a reference solution for interactive recommendation. However, as such algorithms directly interact with users for improved recommendations, serious privacy concerns have been raised regarding its practical use. In this work, we propose a differentially private linear contextual bandit algorithm, via a tree-based mechanism to add Laplace or Gaussian noise to model parameters. Our key insight is that as the model converges during online update, the global sensitivity of its parameters shrinks over time (thus named dynamic global sensitivity). Compared with existing solutions, our dynamic global sensitivity analysis allows us to inject less noise to obtain (ϵ, δ)-differential privacy with added regret caused by noise injection in . We provide a rigorous theoretical analysis over the amount of noise added via dynamic global sensitivity and the corresponding upper regret bound of our proposed algorithm. Experimental results on both synthetic and real-world datasets confirmed the algorithm’s advantage against existing solutions.

Deep neural networks (DNNs) demonstrates significant advantages in improving ranking performance in retrieval tasks. Driven by the recent developments in optimization and generalization of DNNs, learning a neural ranking model online from its interactions with users becomes possible. However, the required exploration for model learning has to be performed in the entire neural network parameter space, which is prohibitively expensive and limits the application of such online solutions in practice.

This paper studies a remarkable property of graphs which is the latent hierarchical grouping of nodes, where each node manifests its membership to a specific group based on the context composed by its neighboring nodes. When modeling the neighborhood structure for graph representation learning, most prior works ignore such latent groups and nodes' membership to different groups, not to mention the hierarchy. Thus, they fall short of delivering a comprehensive understanding of the nodes under different contexts in a graph. In this paper, we propose a novel hierarchical attentive membership model for graph embedding, where the latent memberships for each node are dynamically discovered based on its neighboring context. Both group-level and individual-level attentions are performed when aggregating neighboring states to generate node embeddings. We introduce structural constraints to explicitly regularize the inferred memberships of each node, such that a well-defined hierarchical grouping structure is captured. The proposed model outperformed a set of state-of-the-art graph embedding solutions on node classification and link prediction tasks in a variety of graphs including citation networks and social networks. Qualitative evaluations visualize the learned node embeddings along with the inferred memberships, which proved the concept of membership hierarchy and enables explainable embedding learning in graphs.

Existing online learning to rank (OL2R) solutions are limited to linear models, which are incompetent to capture possible non-linear relations between queries and documents. In this work, to unleash the power of representation learning in OL2R, we propose to directly learn a neural ranking model from users’ implicit feedback (e.g., clicks) collected on the fly. We focus on RankNet and LambdaRank, due to their great empirical success and wide adoption in offline settings, and control the notorious explore-exploit trade-off based on the convergence analysis of neural networks using neural tangent kernel. Specifically, in each round of result serving, exploration is only performed on document pairs where the predicted rank order between the two documents is uncertain; otherwise, the ranker’s predicted order will be followed in result ranking. We prove that under standard assumptions our OL2R solution achieves a gap-dependent upper regret bound of O(log 2(T)), in which the regret is defined on the total number of mis-ordered pairs over T rounds. Comparisons against an extensive set of state-of-the-art OL2R baselines on two public learning to rank benchmark datasets demonstrate the effectiveness of the proposed solution.

Explanations in a recommender system assist users make informed decisions among a set of recommended items. Extensive research attention has been devoted to generate natural language explanations to depict how the recommendations are generated and why the users should pay attention to them. However, due to different limitations of those solutions, e.g., template-based or generation-based, it is hard to make the explanations easily perceivable, reliable, and personalized at the same time.

Graph embedding techniques are pivotal in real-world machine learning tasks that operate on graph-structured data, such as social recommendation and protein structure modeling. Embeddings are mostly performed on the node level for learning representations of each node. Since the formation of a graph is inevitably affected by certain sensitive node attributes, the node embeddings can inherit such sensitive information and introduce undesirable biases in downstream tasks. Most existing works impose ad-hoc constraints on the node embeddings to restrict their distributions for unbiasedness/fairness, which however compromise the utility of the resulting embeddings. In this paper, we propose a principled new way for unbiased graph embedding by learning node embeddings from an underlying bias-free graph, which is not influenced by sensitive node attributes. Motivated by this new perspective, we propose two complementary methods for uncovering such an underlying graph, with the goal of introducing minimum impact on the utility of the embeddings. Both our theoretical justification and extensive experimental comparisons against state-of-the-art solutions demonstrate the effectiveness of our proposed methods.

As recommendation is essentially a comparative (or ranking) process, a good explanation should illustrate to users why an item is believed to be better than another, i.e., comparative explanations about the recommended items. Ideally, after reading the explanations, a user should reach the same ranking of items as the system’s. Unfortunately, little research attention has yet been paid on such comparative explanations.

The rapid development of machine learning on acoustic signal processing has resulted in many solutions for detecting emotions from speech. Early works were developed for clean and acted speech and for a fixed set of emotions. Importantly, the datasets and solutions assumed that a person only exhibited one of these emotions. More recent work has continually been adding realism to emotion detection by considering issues such as reverberation, de-amplification, and background noise, but often considering one dataset at a time, and also assuming all emotions are accounted for in the model. We significantly improve realistic considerations for emotion detection by (i) more comprehensively assessing different situations by combining the five common publicly available datasets as one and enhancing the new dataset with data augmentation that considers reverberation and de-amplification, (ii) incorporating 11 typical home noises into the acoustics, and (iii) considering that in real situations a person may be exhibiting many emotions that are not currently of interest and they should not have to fit into a pre-fixed category nor be improperly labeled. Our novel solution combines CNN with out-of-data distribution detection. Our solution increases the situations where emotions can be effectively detected and outperforms a state-of-the-art baseline.

Crowdsourcing provides a practical way to obtain large amounts of labeled data at a low cost. However, the annotation quality of annotators varies considerably, which imposes new challenges in learning a high-quality model from the crowdsourced annotations. In this work, we provide a new perspective to decompose annotation noise into common noise and individual noise and differentiate the source of confusion based on instance difficulty and annotator expertise on a per-instance-annotator basis. We realize this new crowdsourcing model by an end-to-end learning solution with two types of noise adaptation layers: one is shared across annotators to capture their commonly shared confusions, and the other one is pertaining to each annotator to realize individual confusion. To recognize the source of noise in each annotation, we use an auxiliary network to choose from the two noise adaptation layers with respect to both instances and annotators. Extensive experiments on both synthesized and real-world benchmarks demonstrate the effectiveness of our proposed common noise adaptation solution.

Non-stationary bandits and clustered bandits lift the restrictive assumptions in contextual bandits and provide solutions to many important real-world scenarios. Though they have been studied independently so far, we point out the essence in solving these two problems overlaps considerably. In this work, we connect these two strands of bandit research under the notion of test of homogeneity, which seamlessly addresses change detection for non-stationary bandit and cluster identification for clustered bandit in a unified solution framework. Rigorous regret analysis and extensive empirical evaluations demonstrate the value of our proposed solution, especially its flexibility in handling various environment assumptions, e.g., a clustered non-stationary environment.

Crowdsourcing provides an efficient label collection schema for supervised machine learning. However, to control annotation cost, each instance in the crowdsourced data is typically annotated by a small number of annotators. This creates a sparsity issue and limits the quality of machine learning models trained on such data. In this paper, we study how to handle sparsity in crowdsourced data using data augmentation. Specifically, we propose to directly learn a classifier by augmenting the raw sparse annotations. We implement two principles of high-quality augmentation using Generative Adversarial Networks: 1) the generated annotations should follow the distribution of authentic ones, which is measured by a discriminator; 2) the generated annotations should have high mutual information with the ground-truth labels, which is measured by an auxiliary network. Extensive experiments and comparisons against an array of state-of-the-art learning from crowds methods on three real-world datasets proved the effectiveness of our data augmentation framework. It shows the potential of our algorithm for low-budget crowdsourcing in general.

Sequential recommendation is the task of predicting the next items for users based on their interaction history. Modeling the dependence of the next action on the past actions accurately is crucial to this problem. Moreover, sequential recommendation often faces serious sparsity of item-to-item transitions in a user's action sequence, which limits the practical utility of such solutions.

Search engines are perceived as a reliable source for general information needs. However, finding the answer to medical questions using search engines can be challenging for an ordinary user. Content can be biased and results may present different opinions. In addition, interpreting medically related content can be difficult for users with no medical background. All of these can lead users to incorrect conclusions regarding health related questions.

Collaborative bandit learning, i.e., bandit algorithms that utilize collaborative filtering techniques to improve sample efficiency in online interactive recommendation, has attracted much research attention as it enjoys the best of both worlds. However, all existing collaborative bandit learning solutions impose a stationary assumption about the environment, i.e., both user preferences and the dependency among users are assumed static over time. Unfortunately, this assumption hardly holds in practice due to users' ever-changing interests and dependency relations, which inevitably costs a recommender system sub-optimal performance in practice. In this work, we develop a collaborative dynamic bandit solution to handle a changing environment for recommendation. We explicitly model the underlying changes in both user preferences and their dependency relation as a stochastic process. Individual user's preference is modeled by a mixture of globally shared contextual bandit models with a Dirichlet process prior. Collaboration among users is thus achieved via Bayesian inference over the global bandit models. To balance exploitation and exploration during the interactions, Thompson sampling is used for both model selection and arm selection. Our solution is proved to maintain a standard ~O(√T) Bayesian regret in this challenging environment. Extensive empirical evaluations on both synthetic and real-world datasets further confirmed the necessity of modeling a changing environment and our algorithm's practical advantages against several state-of-the-art online learning solutions.

Information retrieval (IR) in nature is a process of sequential decision making. The system repeatedly interacts with the users to refine its understanding of the users' information needs, improve its estimation of result relevance, and thus increase the utility of its returned results (e.g., the result rankings). Distinct from traditional IR solutions that rigidly execute an offline trained policy, interactive information retrieval emphasizes online policy learning. This, however, is fundamentally difficult for at least three reasons. First, the system only collects user feedback on the presented results, aka, the bandit feedback. Second, users' feedback is known to be noisy and biased. Third, as a result, the system always faces the conflicting goals of improving its policy by presenting currently underestimated results to users versus satisfying the users by ranking the currently estimated best results on top. In this tutorial, we will first motivate the need for online policy learning in interactive IR, by highlighting its importance in several real-world IR problems where online sequential decision making is necessary, such as web search and recommendations. We will carefully address the new challenges that arose in such a solution paradigm, including sample complexity, costly and even outdated feedback, and ethical considerations in online learning (such as fairness and privacy) in interactive IR. We will prepare the technical discussions by first introducing several classical interactive learning strategies from machine learning literature, and then fully dive into the recent research developments for addressing the aforementioned fundamental challenges in interactive IR. Note that the tutorial on "Interactive Information Retrieval: Models, Algorithms, and Evaluation" will provide a broad overview on the general conceptual framework and formal models in interactive IR, while this tutorial covers the online policy learning solutions for interactive IR with bandit feedback.

Recent advances in unbiased learning to rank (LTR) count on Inverse Propensity Scoring (IPS) to eliminate bias in implicit feedback. Though theoretically sound in correcting the bias introduced by treating clicked documents as relevant, IPS ignores the bias caused by (implicitly) treating non-clicked ones as irrelevant. In this work, we first rigorously prove that such use of click data leads to unnecessary pairwise comparisons between relevant documents, which prevent unbiased ranker optimization.

Textual explanations have proved to help improve user satisfaction on machine-made recommendations. However, current mainstream solutions loosely connect the learning of explanation with the learning of recommendation: for example, they are often separately modeled as rating prediction and content generation tasks. In this work, we propose to strengthen their connection by enforcing the idea of sentiment alignment between a recommendation and its corresponding explanation. At training time, the two learning tasks are joined by a latent sentiment vector, which is encoded by the recommendation module and used to make word choices for explanation generation. At both training and inference time, the explanation module is required to generate explanation text that matches sentiment predicted by the recommendation module. Extensive experiments demonstrate our solution outperforms a rich set of baselines in both recommendation and explanation tasks, especially on the improved quality of its generated explanations. More importantly, our user studies confirm our generated explanations help users better recognize the differences between recommended items and understand why an item is recommended.

Online Learning to Rank (OL2R) eliminates the need of explicit relevance annotation by directly optimizing the rankers from their interactions with users. However, the required exploration drives it away from successful practices in offline learning to rank, which limits OL2R’s empirical performance and practical applicability. In this work, we propose to estimate a pairwise learning to rank model online. In each round, candidate documents are partitioned and ranked according to the model’s confidence on the estimated pairwise rank order, and exploration is only performed on the uncertain pairs of documents, i.e., divide-and-conquer. Regret directly defined on the number of mis-ordered pairs is proven, which connects the online solution’s theoretical convergence with its expected ranking performance. Comparisons against an extensive list of OL2R baselines on two public learning to rank benchmark datasets demonstrate the effectiveness of the proposed solution.

Smart Building Technologies hold promise for better livability for residents and lower energy footprints. Yet, the rollout of these technologies, from demand response controls to fault detection and diagnosis, significantly lags behind and is impeded by the current practice of manual identification of sensing point relationships, e.g., how equipment is connected or which sensors are co-located in the same space. This manual process is still error-prone, albeit costly and laborious.We study relation inference among sensor time series. Our key insight is that, as equipment is connected or sensors co-locate in the same physical environment, they are affected by the same real-world events, e.g., a fan turning on or a person entering the room, thus exhibiting correlated changes in their time series data. To this end, we develop a deep metric learning solution that first converts the primitive sensor time series to the frequency domain, and then optimizes a representation of sensors that encodes their relations. Built upon the learned representation, our solution pinpoints the relationships among sensors via solving a combinatorial optimization problem. Extensive experiments on real-world buildings demonstrate the effectiveness of our solution.

A key barrier to applying any smart technology to a building is the requirement of locating and connecting to the necessary resources among the thousands of sensing and control points, i.e., the metadata mapping problem. Existing solutions depend on exhaustive manual annotation of sensor metadata — a laborious, costly, and hardly scalable process. To reduce the amount of manual effort required, this paper presents a multi-oracle selective sampling framework to leverage noisy labels from information sources with unknown reliability such as existing buildings, which we refer to as weak oracles, for metadata mapping. This framework involves an interactive process, where a small set of sensor instances are progressively selected and labeled for it to learn how to aggregate the noisy labels as well as to predict sensor types.Two key challenges arise in designing the framework, namely, weak oracle reliability estimation and instance selection for querying. To address the first challenge, we develop a clustering-based approach for weak oracle reliability estimation to capitalize on the observation that weak oracles perform differently in different groups of instances. For the second challenge, we propose a disagreement-based query selection strategy to combine the potential effect of a labeled instance on both reducing classifier uncertainty and improving the quality of label aggregation. We evaluate our solution on a large collection of real-world building sensor data from 5 buildings with more than 11, 000 sensors of 18 different types. The experiment results validate the effectiveness of our solution, which outperforms a set of state-of-the-art baselines.

Edges play a crucial role in passing information on a graph, especially when they carry textual content reflecting semantics behind how nodes are linked and interacting with each other. In this paper, we propose a channel-aware attention mechanism enabled by edge text content when aggregating information from neighboring nodes; and we realize this mechanism in a graph autoencoder framework. Edge text content is encoded as low-dimensional mixtures of latent topics, which serve as semantic channels for topic-level information passing on edges. We embed nodes and topics in the same latent space to capture their mutual dependency when decoding the structural and textual information on graph. We evaluated the proposed model on Yelp user-item bipartite graph and StackOverflow user-user interaction graph. The proposed model outperformed a set of baselines on link prediction and content prediction tasks. Qualitative evaluations also demonstrated the descriptive power of the learnt node embeddings, showing its potential as an interpretable representation of graphs.

User-provided multi-aspect evaluations manifest users' detailed feedback on the recommended items and enable fine-grained understanding of their preferences. Extensive studies have shown that modeling such data greatly improves the effectiveness and explainability of the recommendations. However, as ranking is essential in recommendation, there is no principled solution yet for collectively generating multiple item rankings over different aspects.

Learning is a predominant theme for any intelligent system, humans, or machines. Moving beyond the classical paradigm of learning from past experience, e.g., offline supervised learning from given labels, a learner needs to actively collect exploratory feedback to learn from the unknowns, i.e., learning through exploration. This tutorial will introduce the learning by exploration paradigm, which is the key ingredient in many interactive online learning problems, including the multi-armed bandit and, more generally, reinforcement learning problems.

Collaborative bandit learning has become an emerging focus for personalized recommendation. It leverages user dependence for joint model estimation and recommendation. As such online learning solutions directly learn from users, e.g., result clicks, they bring in new challenges in privacy protection. Despite the existence of recent studies about privacy in contextual bandit algorithms, how to efficiently protect user privacy in a collaborative bandit learning environment remains unknown.

Family caregivers often report increased anxiety and depression. In order to improve the interactions between in-home patients and caregivers, and reduce strain on caregivers, we build a monitoring, modeling, and interactive recommendation system for caregivers for in-home dementia patient care. The system includes monitoring for mood by speech, building classifiers that work in realistic home settings, and supporting an adaptive recommendation system to reduce stress of the caregiver. This demo shows how our system supports caregivers in practice through several scenarios.

User representation learning is vital to capture diverse user preferences, while it is also challenging as user intents are latent and scattered among complex and different modalities of user-generated data, thus, not directly measurable. Inspired by the concept of user schema in social psychology, we take a new perspective to perform user representation learning by constructing a shared latent space to capture the dependency among different modalities of user-generated data. Both users and topics are embedded to the same space to encode users' social connections and text content, to facilitate joint modeling of different modalities, via a probabilistic generative framework. We evaluated the proposed solution on large collections of Yelp reviews and StackOverflow discussion posts, with their associated network structures. The proposed model outperformed several state-of-the-art topic modeling based user models with better predictive power in unseen documents, and state-of-the-art network embedding based user models with improved link prediction quality in unseen nodes. The learnt user representations are also proved to be useful in content recommendation, e.g., expert finding in StackOverflow.

Predicting users’ preferences based on their sequential behaviors in history is challenging and crucial for modern recommender systems. Most existing sequential recommendation algorithms focus on transitional structure among the sequential actions, but largely ignore the temporal and context information, when modeling the influence of a historical event to current prediction.

Residential homes constitute roughly one-fourth of the total energy usage worldwide. Providing appliance-level energy breakdown has been shown to induce positive behavioral changes that can reduce energy consumption by 15%. Existing approaches for energy breakdown either require hardware installation in every target home or demand a large set of energy sensor data available for model training. However, very few homes in the world have installed sub-meters (sensors measuring individual appliance energy); and the cost of retrofitting a home with extensive sub-metering eats into the funds available for energy saving retrofits. As a result, strategically deploying sensing hardware to maximize the reconstruction accuracy of sub-metered readings in non-instrumented homes while minimizing deployment costs becomes necessary and promising. In this work, we develop an active learning solution based on low-rank tensor completion for energy breakdown. We propose to actively deploy energy sensors to appliances from selected homes, with a goal to improve the prediction accuracy of the completed tensor with minimum sensor deployment cost. We empirically evaluate our approach on the largest public energy dataset collected in Austin, Texas, USA, from 2013 to 2017. The results show that our approach gives better performance with fixed number of sensors installed, when compared to the state-of-the-art, which is also proven by our theoretical analysis.

Huazheng Wang, Zhe Gan, Xiaodong Liu, Jingjing Liu, Jianfeng Gao, Hongning Wang. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

In this paper, we study the problem of online influence maximization in social networks. In this problem, a learner aims to identify the set of "best influencers" in a network by interacting with the network, i.e., repeatedly selecting seed nodes and observing activation feedback in the network. We capitalize on an important property of the influence maximization problem named network assortativity, which is ignored by most existing works in online influence maximization. To realize network assortativity, we factorize the activation probability on the edges into latent factors on the corresponding nodes, including influence factor on the giving nodes and susceptibility factor on the receiving nodes. We propose an upper confidence bound based online learning solution to estimate the latent factors, and therefore the activation probabilities. Considerable regret reduction is achieved by our factorization based online influence maximization algorithm. Extensive empirical evaluations on two real-world networks showed the effectiveness of our proposed solution.

Reinforcement learning is effective in optimizing policies for recommender systems. Current solutions mostly focus on model-free approaches, which require frequent interactions with a real environment, and thus are expensive in model learning. Offline evaluation methods, such as importance sampling, can alleviate such limitations, but usually request a large amount of logged data and do not work well when the action space is large. In this work, we propose a model-based reinforcement learning solution which models the user-agent interaction for offline policy learning via a generative adversarial network. To reduce bias in the learnt policy, we use the discriminator to evaluate the quality of generated sequences and rescale the generated rewards. Our theoretical analysis and empirical evaluations demonstrate the effectiveness of our solution in identifying patterns from given offline data and learning policies based on the offline and generated data.

We present a context-aware neural ranking model to exploit users' on-task search activities and enhance retrieval performance. In particular, a two-level hierarchical recurrent neural network is introduced to learn search context representation of individual queries, search tasks, and corresponding dependency structure by jointly optimizing two companion retrieval tasks: document ranking and query suggestion. To identify variable dependency structure between search context and users' ongoing search activities, attention at both levels of recurrent states are introduced. Extensive experiment comparisons against a rich set of baseline methods and an in-depth ablation analysis confirm the value of our proposed approach for modeling search context buried in search tasks.

Latent factor models have achieved great success in personalized recommendations, but they are also notoriously difficult to explain. In this work, we integrate regression trees to guide the learning of latent factor models for recommendation, and use the learnt tree structure to explain the resulting latent factors. Specifically, we build regression trees on users and items respectively with user-generated reviews, and associate a latent profile to each node on the trees to represent users and items. With the growth of regression tree, the latent factors are gradually refined under the regularization imposed by the tree structure. As a result, we are able to track the creation of latent profiles by looking into the path of each factor on regression trees, which thus serves as an explanation for the resulting recommendations. Extensive experiments on two large collections of Amazon and Yelp reviews demonstrate the advantage of our model over several competitive baseline algorithms. Besides, our extensive user study also confirms the practical value of explainable recommendations generated by our model.

Online Learning to Rank (OL2R) algorithms learn from implicit user feedback on the fly. The key to such algorithms is an unbiased estimate of gradients, which is often (trivially) achieved by uniformly sampling from the entire parameter space. Unfortunately, this leads to high-variance in gradient estimation, resulting in high regret during model updates, especially when the dimension of the parameter space is large.

A user-generated review document is a product between the item's intrinsic properties and the user's perceived composition of those properties. Without properly modeling and decoupling these two factors, one can hardly obtain any accurate user understanding nor item profiling from such user-generated data. In this paper, we study a new text mining problem that aims at differentiating a user's subjective composition of topical content in his/her review document from the entity's intrinsic properties. Motivated by the Item Response Theory (IRT), we model each review document as a user's detailed response to an item, and assume the response is jointly determined by the individuality of the user and the property of the item. We model the text-based response with a generative topic model, in which we characterize the items' properties and users' manifestations of them in a low-dimensional topic space. Via posterior inference, we separate and study these two components over a collection of review documents. Extensive experiments on two large collections of Amazon and Yelp review data verified the effectiveness of the proposed solution: it outperforms the state-of-art topic models with better predictive power in unseen documents, which is directly translated into improved performance in item recommendation and item summarization tasks.

research-article Share on A Tree-Structured Neural Network Model for Household Energy Breakdown Authors: Yiling Jia University of Virginia, USA University of Virginia, USAView Profile , Nipun Batra IIT Gandhinagar, India IIT Gandhinagar, IndiaView Profile , Hongning Wang University of Virginia, USA University of Virginia, USAView Profile , Kamin Whitehouse University of Virginia, USA University of Virginia, USAView Profile Authors Info & Claims WWW '19: The World Wide Web ConferenceMay 2019 Pages 2872–2878https://doi.org/10.1145/3308558.3313405Published:13 May 2019Publication History 14citation306DownloadsMetricsTotal Citations14Total Downloads306Last 12 Months61Last 6 weeks6 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.

Recommender systems have to handle a highly non-stationary environment, due to users' fast changing interests over time. Traditional solutions have to periodically rebuild their models, despite high computational cost. But this still cannot empower them to automatically adjust to abrupt changes in trends caused by timely information. It is important to note that the changes of reward distributions caused by a non-stationary environment can also be context dependent. When the change is orthogonal to the given context, previously maintained models should be reused for better recommendation prediction.

Homes constitute roughly one-third of the total energy usage worldwide. Providing an energy breakdown – energy consumption per appliance, can help save up to 15% energy. Given the vast differences in energy consumption patterns across different regions, existing energy breakdown solutions require instrumentation and model training for each geographical region, which is prohibitively expensive and limits the scalability. In this paper, we propose a novel region independent energy breakdown model via statistical transfer learning. Our key intuition is that the heterogeneity in homes and weather across different regions most significantly impacts the energy consumption across regions; and if we can factor out such heterogeneity, we can learn region independent models or the homogeneous energy breakdown components for each individual appliance. Thus, the model learnt in one region can be transferred to another region. We evaluate our approach on two U.S. cities having distinct weather from a publicly available dataset. We find that our approach gives better energy breakdown estimates requiring the least amount of instrumented homes from the target region, when compared to the state-of-the-art.

The massively available data about user engagement with online information service systems provides a gold mine about users' latent intents. It calls for quantitative user behavior modeling. In this paper, we study the problem by looking into users' sequential interactive behaviors. Inspired by the concepts of episodic memory and semantic memory in cognitive psychology, which describe how users' behaviors are differently influenced by past experience, we propose a Long- and Short-term Hawkes Process model. It models the short-term dependency between users' actions within a period of time via a multi-dimensional Hawkes process and the long-term dependency between actions across different periods of time via a one dimensional Hawkes process. Experiments on two real-world user activity log datasets (one from an e-commerce website and one from a MOOC website) demonstrate the effectiveness of our model in capturing the temporal dependency between actions in a sequence of user behaviors. It directly leads to improved accuracy in predicting the type and the time of the next action. Interestingly, the inferred dependency between actions in a sequence sheds light on the underlying user intent behind direct observations and provides insights for downstream applications.

User modeling is critical for understanding user intents, while it is also challenging as user intents are so diverse and not directly observable. Most existing works exploit specific types of behavior signals for user modeling, e.g., opinionated data or network structure; but the dependency among different types of user-generated data is neglected.

Part-of-speech (POS) tagging is the foundation of many natural language processing applications. Rule-based POS tagging is a wellknown solution, which assigns tags to the words using a set of predefined rules. Many researchers favor statistical-based approaches over rule-based methods for better empirical accuracy. However, until now, the computational cost of rule-based POS tagging has made it difficult to study whether more complex rules or larger rulesets could lead to accuracy competitive with statistical approaches. In this paper, we leverage two hardware accelerators, the Automata Processor (AP) and Field Programmable Gate Arrays (FPGA), to accelerate rule-based POS tagging by converting rules to regular expressions and exploiting the highly-parallel regular-expressionmatching ability of these accelerators. We study the relationship between rule set size and accuracy, and observe that adding more rules only poses minimal overhead on the AP and FPGA. This allows a substantial increase in the number and complexity of rules, leading to accuracy improvement. Our experiments on Treebank and Brown corpora achieve up to 2,600X and 1,914X speedups on the AP and on the FPGA respectively over rule-based methods on the CPU in the rule-matching stage, up to 58× speedup over the Perceptron POS tagger on the CPU in total testing time, and up to 253× speedup over the LSTM tagger on the GPU in total testing time, while showing a competitive accuracy compared to neural-network and statistical solutions.

Implicit feedback, such as user clicks, although abundant in online information service systems, does not provide substantial evidence on users' evaluation of system's output. Without proper modeling, such incomplete supervision inevitably misleads model estimation, especially in a bandit learning setting where the feedback is acquired on the fly. In this work, we perform contextual bandit learning with implicit feedback by modeling the feedback as a composition of user result examination and relevance judgment. Since users' examination behavior is unobserved, we introduce latent variables to model it. We perform Thompson sampling on top of variational Bayesian inference for arm selection and model update. Our upper regret bound analysis of the proposed algorithm proves its feasibility of learning from implicit feedback in a bandit setting; and extensive empirical evaluations on click logs collected from a major MOOC platform further demonstrate its learning effectiveness in practice.

Modern web search engines exploit users' search history to personalize search results, with a goal of improving their service utility on a per-user basis. But it is this very dimension that leads to the risk of privacy infringement and raises serious public concerns. In this work, we propose a client-centered intent-aware query obfuscation solution for protecting user privacy in a personalized web search scenario. In our solution, each user query is submitted with l additional cover queries and corresponding clicks, which act as decoys to mask users' genuine search intent from a search engine. The cover queries are sequentially sampled from a set of hierarchically organized language models to ensure the coherency of fake search intents in a cover search task. Our approach emphasizes the plausibility of generated cover queries, not only to the current genuine query but also to previous queries in the same task, to increase the complexity for a search engine to identify a user's true intent. We also develop two new metrics from an information theoretic perspective to evaluate the effectiveness of provided privacy protection. Comprehensive experiment comparisons with state-of-the-art query obfuscation techniques are performed on the public AOL search log, and the propitious results substantiate the effectiveness of our solution.

Online learning to rank (OL2R) optimizes the utility of returned search results based on implicit feedback gathered directly from users. In this paper, we accelerate the online learning process by efficient exploration in the gradient space. Our algorithm, named as Null Space Gradient Descent, reduces the exploration space to only the null space of recent poorly performing gradients. This prevents the algorithm from repeatedly exploring directions that have been discouraged by the most recent interactions with users. To improve sensitivity of the resulting interleaved test, we selectively construct candidate rankers to maximize the chance that they can be differentiated by candidate ranking documents in the current query; and we use historically difficult queries to identify the best ranker when tie occurs in comparing the rankers. Extensive experimental comparisons with the state-of-the-art OL2R algorithms on several public benchmarks confirmed the effectiveness of our proposal algorithm, especially in its fast learning convergence and promising ranking quality at an early stage.

Explaining automatically generated recommendations allows users to make more informed and accurate decisions about which results to utilize, and therefore improves their satisfaction. In this work, we develop a multi-task learning solution for explainable recommendation. Two companion learning tasks of user preference modeling for recommendation and opinionated content modeling for explanation are integrated via a joint tensor factorization. As a result, the algorithm predicts not only a user's preference over a list of items, i.e., recommendation, but also how the user would appreciate a particular item at the feature level, i.e., opinionated textual explanation. Extensive experiments on two large collections of Amazon and Yelp reviews confirmed the effectiveness of our solution in both recommendation and explanation tasks, compared with several existing recommendation algorithms. And our extensive user study clearly demonstrates the practical value of the explainable recommendations generated by our algorithm.

Multi-armed bandit algorithms have become a reference solution for handling the explore/exploit dilemma in recommender systems, and many other important real-world problems, such as display advertisement. However, such algorithms usually assume a stationary reward distribution, which hardly holds in practice as users' preferences are dynamic. This inevitably costs a recommender system consistent suboptimal performance. In this paper, we consider the situation where the underlying distribution of reward remains unchanged over (possibly short) epochs and shifts at unknown time instants. In accordance, we propose a contextual bandit algorithm that detects possible changes of environment based on its reward estimation confidence and updates its arm selection strategy respectively. Rigorous upper regret bound analysis of the proposed algorithm demonstrates its learning effectiveness in such a non-trivial environment. Extensive empirical evaluations on both synthetic and real-world datasets for recommendation confirm its practical utility in a changing environment.

We develop Hide-n-Seek, an intent-aware privacy protection plugin for personalized web search. In addition to users' genuine search queries, Hide-n-Seek submits k cover queries and corresponding clicks to an external search engine to disguise a user's search intent grounded and reinforced in a search session by mimicking the true query sequence. The cover queries are synthesized and randomly sampled from a topic hierarchy, where each node represents a coherent search topic estimated by both n-gram and neural language models constructed over crawled web documents. Hide-n-Seek also personalizes the returned search results by re-ranking them based on the genuine user profile developed and maintained on the client side. With a variety of graphical user interfaces, we present the topic-based query obfuscation mechanism to the end users for them to digest how their search privacy is protected.

Homes constitute more than one-thirds of the total energy consumption. Producing an energy breakdown for a home has been shown to reduce household energy consumption by up to 15%, among other benefits. However, existing approaches to produce an energy breakdown require hardware to be installed in each home and are thus prohibitively expensive. In this paper, we propose a novel application of feature-based matrix factorisation that does not require any additional hard- ware installation. The basic premise of our approach is that common design and construction patterns for homes create a repeating structure in their energy data. Thus, a sparse basis can be used to represent energy data from a broad range of homes. We evaluate our approach on 516 homes from a publicly available data set and find it to be more effective than five baseline approaches that either require sensing in each home, or a very rigorous survey across a large number of homes coupled with complex modelling. We also present a deployment of our system as a live web application that can potentially provide energy breakdown to millions of homes.

We perform online interactive recommendation via a factorization-based bandit algorithm. Low-rank matrix completion is performed over an incrementally constructed user-item preference matrix, where an upper confidence bound based item selection strategy is developed to balance the exploit/explore trade-off during online learning. Observable contextual features and dependency among users (e.g., social influence) are leveraged to improve the algorithm's convergence rate and help conquer cold-start in recommendation. A high probability sublinear upper regret bound is proved for the developed algorithm, where considerable regret reduction is achieved on both user and item sides. Extensive experimentations on both simulations and large-scale real-world datasets confirmed the advantages of the proposed algorithm compared with several state-of-the-art factorization-based and bandit-based collaborative filtering methods.

The recorded student activities in Massive Open Online Course (MOOC) provide us a unique opportunity to model their learning behaviors, identify their particular learning intents, and enable personalized assistance and guidance in online education. In this work, based on a thorough qualitative study of students' behaviors recorded in two MOOC courses with large student enrollments, we develop a non-parametric Bayesian model to capture students' sequential learning activities in a generative manner. Homogeneity of students' learning behaviors is captured by clustering them into latent student groups, where shared model structure characterizes the transitional patterns, intensity and temporal distribution of their learning activities. In the meanwhile, heterogeneity is captured by clustering students into different groups. Both qualitative and quantitative studies on those two MOOC courses confirmed the effectiveness of the proposed model in identifying students' learning behavior patterns and clustering them into related groups for predictive analysis. The identified student groups accurately predict student retention, course satisfaction and demographics.

In this work, we propose to improve long-term user engagement in a recommender system from the perspective of sequential decision optimization, where users' click and return behaviors are directly modeled for online optimization. A bandit-based solution is formulated to balance three competing factors during online learning, including exploitation for immediate click, exploitation for expected future clicks, and exploration of unknowns for model estimation. We rigorously prove that with a high probability our proposed solution achieves a sublinear upper regret bound in maximizing cumulative clicks from a population of users in a given period of time, while a linear regret is inevitable if a user's temporal return behavior is not considered when making the recommendations. Extensive experimentation on both simulations and a large-scale real-world dataset collected from Yahoo frontpage news recommendation log verified the effectiveness and significant improvement of our proposed algorithm compared with several state-of-the-art online learning baselines for recommendation.

Textual health advice generated from different online sources (e.g., health apps and websites) can be conflicting. Conflicts can occur due to lexical features, (such as, negation, antonyms, or numerical mismatch) or can be conditioned upon time and/or physiological status. Detecting conflicts from textual health advice poses several challenges, including, large structural variation between text and hypothesis pairs, finding conceptual overlap between pairs of advice, and inference of the semantics of an advice (i.e., what to do, why, and how). In this demonstration, we present a semantic rule-based system to detect different types of conflicts in online textual health advice statements in a context-aware and interpretable manner.

With the rapid digitalization of the health sector, people often turn to mobile apps and online health websites for health advice. Health advice generated from different sources can be conflicting as they address different aspects of health (e.g., weight loss, diet, disease) or as they are unaware of the context of a user (e.g., age, gender, physiological condition). Conflicts can occur due to lexical features, (such as, negation, antonyms, or numerical mismatch) or can be conditioned upon time and/or physiological status. We formulate the problem of finding conflicting health advice and develop a comprehensive taxonomy of conflicts. While a similar research area in the natural language processing domain explores the problem of textual contradiction identification, finding conflicts in health advice poses its own unique lexical and semantic challenges. These include large structural variation between text and hypothesis pairs, finding conceptual overlap between pairs of advice, and inference of the semantics of an advice (i.e., what to do, why and how). Hence, we develop Preclude, a novel semantic rule-based solution to detect conflicting health advice derived from heterogeneous sources utilizing linguistic rules and external knowledge bases. As our solution is interpretable and comprehensive, it can guide users towards conflict resolution too. We evaluate Preclude using 1156 real advice statements covering 8 important health topics that are collected from smart phone health apps and popular health websites. Preclude results in 90% accuracy and outperforms the accuracy and F1 score of the baseline approach by about 1.5 times and 3 times, respectively.

One important way for people to make their voice heard is to comment on the articles they have read online, such as news reports and each other's posts. The user-generated comments together with the commented documents form a unique correspondence structure. Properly modeling the dependency in such data is thus vital for one to obtain accurate insight of people's opinions and attention.

We develop an aspect-based sentiment analysis system named ReviewMiner. It analyzes opinions expressed about an entity in an online review at the level of topical aspects to discover each individual reviewer's latent opinion on each aspect as well as his/her relative emphasis on different aspects when forming the overall judgment of the entity. The system personalizes the retrieved results according to users' input preferences over the identified aspects, recommends similar items based on the detailed aspect-level opinions, and summarizes aspect-level opinions in textual, temporal and spatial dimensions. The unique multi-modal opinion summarization and visualization mechanisms provide users with rich perspectives to digest information from user-generated opinionated content for making informed decisions.

We propose to capture humans' variable and idiosyncratic sentiment via building personalized sentiment classification models at a group level. Our solution roots in the social comparison theory that humans tend to form groups with others of similar minds and ability, and the cognitive consistency theory that mutual influence inside groups will eventually shape group norms and attitudes, with which group members will all shift to align. We formalize personalized sentiment classification as a multi-task learning problem. In particular, to exploit the clustering property of users' opinions, we impose a non-parametric Dirichlet Process prior over the personalized models, in which group members share the same customized sentiment model adapted from a global classifier. Extensive experimental evaluations on large collections of Amazon and Yelp reviews confirm the effectiveness of the proposed solution: it outperformed user-independent classification solutions, and several state-of-the-art model adaptation and multi-task learning algorithms.

In recent years, mobile devices have become the most popular interface for users to retrieve and access information: recent reports show that users spend significantly more time and issue more search queries on mobile devices than on desktops in the United States. 1 The accelerated growth of mobile usage brings unique opportunities to the information retrieval and data mining research communities. Mobile devices capture rich contextual and personal signals that can be leveraged to accurately predict users’ intent for serving more relevant content and can even proactively provide novel zero-query recommendations. Apple Siri, Google Now, and Microsoft Cortana are recent examples of such emerging systems. Furthermore, mobile devices constantly generate a huge amount of sensor footprints (e.g., GPS, motion sensors) and user activity data (e.g., used apps) that are often missing from their desktop counterparts. These new sources of implicit and explicit user feedback are valuable for discovering actionable knowledge, and designing better systems that serve each individual the right content at the right time and location. In addition, by aggregating mobile interactions across individuals, one can infer interesting conclusions beyond search and recommendation. Generating real-time traffic estimates is one example of such applications. This special issue focuses on research problems of search, mining, and their applications in mobile devices. Topics of interest in this special issue include but are not limited to mobile data mining and management, mobile search, personalization and recommendation, mobile user interfaces and human-computer interaction, and new applications in the mobile environment. The aim of this special issue is to bring together top experts across multiple disciplines, including information retrieval, data mining, mobile computing, and cyberphysical systems, such that academic and industrial researchers can exchange ideas and share the latest developments on the state of the art and practice of mobile search and mobile data mining.

Motivated by the findings in social science that people's opinions are diverse and variable while together they are shaped by evolving social norms, we perform personalized sentiment classification via shared model adaptation over time.In our proposed solution, a global sentiment model is constantly updated to capture the homogeneity in which users express opinions, while personalized models are simultaneously adapted from the global model to recognize the heterogeneity of opinions from individuals.Global model sharing alleviates data sparsity issue, and individualized model adaptation enables efficient online model learning.Extensive experimentations are performed on two large review collections from Amazon and Yelp, and encouraging performance gain is achieved against several state-of-the-art transfer learning and multi-task learning based sentiment classification solutions.

Contextual bandit algorithms provide principled online learning solutions to find optimal trade-offs between exploration and exploitation with companion side-information. Most contextual bandit algorithms simply assume the learner would have access to the entire set of features, which govern the generation of payoffs from a user to an item. However, in practice it is challenging to exhaust all relevant features ahead of time, and oftentimes due to privacy or sampling constraints many factors are unobservable to the algorithm. Failing to model such hidden factors leads a system to make constantly suboptimal predictions. In this paper, we propose to learn the hidden features for contextual bandit algorithms. Hidden features are explicitly introduced in our reward generation assumption, in addition to the observable contextual features. A scalable bandit algorithm is achieved via coordinate descent, in which closed form solutions exist at each iteration for both hidden features and bandit parameters. Most importantly, we rigorously prove that the developed contextual bandit algorithm achieves a sublinear upper regret bound with high probability, and a linear regret is inevitable if one fails to model such hidden features. Extensive experimentation on both simulations and large-scale real-world datasets verified the advantages of the proposed algorithm compared with several state-of-the-art contextual bandit algorithms and existing ad-hoc combinations between bandit algorithms and matrix factorization methods.

Modern search engines utilize users' search history for personalization, which provides more effective, useful and relevant search results. However, it also has the potential risk of revealing users' privacy by identifying their underlying intention from their logged search behaviors. To address this privacy issue, we proposed a Topic-based Privacy Protection solution on client side. In our solution, each user query will be submitted with k additional cover queries, which will act as a proxy to disguise users' intent from a search engine. The set of cover queries are generated in a controlled way so that each query carries similar uncertainty to randomize a user's search history while still providing necessary utility for the search engine to perform personalization. We used statistical topic models to infer topics from the original user query and generated cover queries of similar entropy but from unrelated topics. Extensive experiments are performed on AOL search log and the promising results demonstrated the effectiveness of our solution.

Contextual bandit algorithms provide principled online learning solutions to find optimal trade-offs between exploration and exploitation with companion side-information. They have been extensively used in many important practical scenarios, such as display advertising and content recommendation. A common practice estimates the unknown bandit parameters pertaining to each user independently. This unfortunately ignores dependency among users and thus leads to suboptimal solutions, especially for the applications that have strong social components.

Various topic models have been developed for sentiment analysis tasks. But the simple topic-sentiment mixture assumption prohibits them from finding fine-grained dependency between topical aspects and sentiments. In this paper, we build a Hidden Topic Sentiment Model (HTSM) to explicitly capture topic coherence and sentiment consistency in an opinionated text document to accurately extract latent aspects and corresponding sentiment polarities. In HTSM, 1) topic coherence is achieved by enforcing words in the same sentence to share the same topic assignment and modeling topic transition between successive sentences; 2) sentiment consistency is imposed by constraining topic transitions via tracking sentiment changes; and 3) both topic transition and sentiment transition are guided by a parameterized logistic function based on the linguistic signals directly observable in a document. Extensive experiments on four categories of product reviews from both Amazon and NewEgg validate the effectiveness of the proposed model.

MOTIVATION: Medical Subject Headings (MeSH) indexing, which is to assign a set of MeSH main headings to citations, is crucial for many important tasks in biomedical text mining and information retrieval. Large-scale MeSH indexing has two challenging aspects: the citation side and MeSH side. For the citation side, all existing methods, including Medical Text Indexer (MTI) by National Library of Medicine and the state-of-the-art method, MeSHLabeler, deal with text by bag-of-words, which cannot capture semantic and context-dependent information well. METHODS: We propose DeepMeSH that incorporates deep semantic information for large-scale MeSH indexing. It addresses the two challenges in both citation and MeSH sides. The citation side challenge is solved by a new deep semantic representation, D2V-TFIDF, which concatenates both sparse and dense semantic representations. The MeSH side challenge is solved by using the 'learning to rank' framework of MeSHLabeler, which integrates various types of evidence generated from the new semantic representation. RESULTS: DeepMeSH achieved a Micro F-measure of 0.6323, 2% higher than 0.6218 of MeSHLabeler and 12% higher than 0.5637 of MTI, for BioASQ3 challenge data with 6000 citations. AVAILABILITY AND IMPLEMENTATION: The software is available upon request. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Commercial and industrial buildings account for a considerable portion of all energy consumed in the U.S., and thus reducing this energy consumption is a national grand challenge. Based on the large deployment of sensors in modern commercial buildings, many organizations are applying data analytic solutions to the thousands of sensing and control points to detect wasteful and incorrect operations for energy savings. Scaling this approach is challenging, however, because the metadata about these sensing and control points is inconsistent between buildings, or even missing altogether. Moreover, normalizing the metadata requires significant integration effort.

Many elderly who are suffering from dementia are also suffering from agitation. While most assisted living facilities and home health care situations rely upon caregivers to monitor and record agitation of their patients, the accuracy is limited because the caregiver must be present during the agitation and must record the event properly. Accurate 24-7 data would help physicians with improved diagnoses and care. To solve this problem we developed KinVocal, a system that continuously monitors and detects agitated vocal events and can be used for the elderly population suffering from dementia. KinVocal, using a novel combination of acoustic signal processing and multiple text mining techniques, automatically detects and records the 8 major vocal agitations for dementia patients as defined by the medical community. This includes: constant unwarranted request for attention or help, making verbal sexual advances, crying, screaming, laughing, cursing, speaking in repetitive sentences, and negativism. The novelty of KinVocal includes the comprehensiveness of addressing all 8 vocal events, using the text of the vocalizations only when accurate, combining text and acoustic features when necessary, and employing text mining and feature identification. A comprehensive performance evaluation includes using data from Youtube and movies, controlled experiments, and real in-home deployments. The results show high accuracy for all 8 vocal events.

Query auto-completion (QAC) facilitates faster user query input by predicting users' intended queries. Most QAC algorithms take a learning-based approach to incorporate various signals for query relevance prediction. However, such models are trained on simulat- ed user inputs from query log data. The lack of real user interaction data in the QAC process prevents them from further improving the QAC performance. In this work, for the first time we collect a high-resolution QAC query log that records every keystroke in a QAC session. Based on this data, we discover two user behaviors, namely the horizontal skipping bias and vertical position bias which are crucial for rele- vance prediction in QAC. In order to better explain them, we pro- pose a novel two-dimensional click model for modeling the QAC process with emphasis on these behaviors. Extensive experiments on our QAC data set from both PC and mobile devices demonstrate that our proposed model can accurate- ly explain the users' behaviors in interacting with a QAC system, and the resulting relevance model significant improves the QAC performance over existing click models. Furthermore, the learned knowledge about the skipping behavior can be effectively incorpo- rated into existing learning-based models to further improve their performance.

Search satisfaction is a property of a user's search process. Understanding it is critical for search providers to evaluate the performance and improve the effectiveness of search engines. Existing methods model search satisfaction holistically at the search-task level, ignoring important dependencies between action-level satisfaction and overall task satisfaction. We hypothesize that searchers' latent action-level satisfaction (i.e., whether they believe they were satisfied with the results of a query or click) influences their observed search behaviors and contributes to overall search satisfaction. We conjecture that by modeling search satisfaction at the action level, we can build more complete and more accurate predictors of search-task satisfaction. To do this, we develop a latent structural learning method, whereby rich structured features and dependency relations unique to search satisfaction prediction are explored. Using in-situ search satisfaction judgments provided by searchers, we show that there is significant value in modeling action-level satisfaction in search-task satisfaction prediction. In addition, experimental results on large-scale logs from Bing.com demonstrate clear benefit from using inferred action satisfaction labels for other applications such as document relevance estimation and query suggestion.

RankNet is one of the widely adopted ranking models for web search tasks. However, adapting a generic RankNet for personalized search is little studied. In this paper, we first continue-trained a variety of RankNets with different number of hidden layers and network structures over a previously trained global RankNet model, and observed that a deep neural network with five hidden layers gives the best performance. To further improve the performance of adaptation, we propose a set of novel methods categorized into two groups. In the first group, three methods are proposed to properly assess the usefulness of each adaptation instance and only leverage the most informative instances to adapt a user-specific RankNet model. These assessments are based on KL-divergence, click entropy or a heuristic to ignore top clicks in adaptation queries. In the second group, two methods are proposed to regularize the training of the neural network in RankNet: one of these methods regularize the error back-propagation via a truncated gradient approach, while the other method limits the depth of the back propagation when adapting the neural network. We empirically evaluate our approaches using a large-scale real-world data set. Experimental results exhibit that our methods all give significant improvements over a strong baseline ranking system, and the truncated gradient approach gives the best performance, significantly better than all others.

Searchers' information needs are diverse and cover a broad range of topics; hence, it is important for search engines to accurately understand each individual user's search intents in order to provide optimal search results. Search log data, which records users' search behaviors when interacting with search engines, provides a valuable source of information about users' search intents. Therefore, properly characterizing the heterogeneity among the users' observed search behaviors is the key to accurately understanding their search intents and to further predicting their behaviors.

In this paper, we study the task of relational entity search which aims at automatically learning an entity ranking function for a desired relation. To rank entities, we exploit the redundancy abound in their snippets; however, such redundancy is noisy as not all the snippets represent information relevant to the desired relation. To explore useful information from such noisy redundancy, we abstract the task as a distantly supervised ranking problem — based on coarse entity-level annotations, deriving a relation-specific ranking function for the purpose of online searching. As the key challenge, without detailed snippet-level annotations, we have to learn an entity ranking function that can effectively filter noise; furthermore, the ranking function should also be online executable. We develop Pattern-based Filter Network (PFNet), a novel probabilistic graphical model, as our solution. To balance the accuracy and efficiency requirements, PFNet selects a limited size of indicative patterns to filter noisy snippets, and inverted indexes are utilized to retrieve required features. Experiments on the large scale CuleWeb09 data set for six different relations confirm the effectiveness of the proposed PFNet model, which outperforms five state-of-the-art relational entity ranking methods.

In typical studies of node grouping detection, the grouping is presumed to have a certain type of correlation with the network structure (e.g., densely connected groups of nodes that are loosely connected in between). People have defined different fitness measures (modularity, conductance, etc.) to quantify such correlation, and group the nodes by optimizing a certain fitness measure. However, a particular grouping with desired semantics, as the target of the detection, is not promised to be detectable by each measure. We study a fundamental problem in the process of node grouping discovery: Given a particular grouping in a network, whether and to what extent it can be discovered with a given fitness measure. We propose two approaches of testing the detectability, namely ranking-based and correlation-based randomization tests. Our methods are evaluated on both synthetic and real datasets, which shows the proposed methods can effectively predict the detectability of groupings of various types, and support explorative process of node grouping discovery.

With the development of social media and social networks, user-generated content, like forums, blogs and comments, are not only getting richer, but also ubiquitously interconnected with many other objects and entities, forming a heterogeneous information network between them. Sentiment analysis on such kinds of data can no longer ignore the information network, since it carries a lot of rich and valuable information, explicitly or implicitly, where some of them can be observed while others are not. In this paper, we propose a novel information network-based framework which can infer hidden similarity and dissimilarity between users by exploring similar and opposite opinions, so as to improve post-level and user-level sentiment classification in the same time. More specifically, we develop a new meta path-based measure for inferring pseudo-friendship as well as dissimilarity between users, and propose a semi-supervised refining model by encoding similarity and dissimilarity from both user-level and post-level relations. We extensively evaluate the proposed approach and compare with several state-of-the-art techniques on two real-world forum datasets. Experimental results show that our proposed model with 10.5% labeled samples can achieve better performance than a traditional supervised model trained on 61.7% data samples.

Search engines train and apply a single ranking model across all users, but searchers' information needs are diverse and cover a broad range of topics. Hence, a single user-independent ranking model is insufficient to satisfy different users' result preferences. Conventional personalization methods learn separate models of user interests and use those to re-rank the results from the generic model. Those methods require significant user history information to learn user preferences, have low coverage in the case of memory-based methods that learn direct associations between query-URL pairs, and have limited opportunity to markedly affect the ranking given that they only re-order top-ranked items.

In this paper, we present a log-based study on user search behavior comparisons on three different platforms: desktop, mobile and tablet. We use three-month search logs in 2012 from a commercial search engine for our study. Our objective is to better understand how and to what extent mobile and tablet searchers behave differently than desktop users. Our study spans a variety of aspects including query categorization, query length, search time distribution, search location distribution, user click patterns and so on. From our data set, we reveal that there are significant differences between user search patterns in these three platforms, and therefore use the same ranking system is not an optimal solution for all of them. Consequently, we propose a framework that leverages a set of domain-specific features, along with the training data from desktop search, to further improve the search relevance for mobile and tablet platforms. Experimental results demonstrate that by transferring knowledge from desktop search, search relevance on mobile and tablet can be greatly improved.

Search tasks, comprising a series of search queries serving the same information need, have recently been recognized as an accurate atomic unit for modeling user search intent. Most prior research in this area has focused on short-term search tasks within a single search session, and heavily depend on human annotations for supervised classification model learning. In this work, we target the identification of long-term, or cross-session, search tasks (transcending session boundaries) by investigating inter-query dependencies learned from users' searching behaviors. A semi-supervised clustering model is proposed based on the latent structural SVM framework, and a set of effective automatic annotation rules are proposed as weak supervision to release the burden of manual annotation. Experimental results based on a large-scale search log collected from Bing.com confirms the effectiveness of the proposed model in identifying cross-session search tasks and the utility of the introduced weak supervision signals. Our learned model enables a more comprehensive understanding of users' search behaviors via search logs and facilitates the development of dedicated search-engine support for long-term tasks.

Click models aim at extracting intrinsic relevance of documents to queries from biased user clicks. One basic modeling assumption made in existing work is to treat such intrinsic relevance as an atomic query-document-specific parameter, which is solely estimated from historical clicks without using any content information about a document or relationship among the clicked/skipped documents under the same query. Due to this overly simplified assumption, existing click models can neither fully explore the information about a document's relevance quality nor make predictions of relevance for any unseen documents.

Personalized search systems tailor search results to the current user intent using historic search interactions. This relies on being able to find pertinent information in that user's search history, which can be challenging for unseen queries and for new search scenarios. Building richer models of users' current and historic search tasks can help improve the likelihood of finding relevant content and enhance the relevance and coverage of personalization methods. The task-based approach can be applied to the current user's search history, or as we focus on here, all users' search histories as so-called "groupization" (a variant of personalization whereby other users' profiles can be used to personalize the search experience). We describe a method whereby we mine historic search-engine logs to find other users performing similar tasks to the current user and leverage their on-task behavior to identify Web pages to promote in the current ranking. We investigate the effectiveness of this approach versus query-based matching and finding related historic activity from the current user (i.e., group versus individual). As part of our studies we also explore the use of the on-task behavior of particular user cohorts, such as people who are expert in the topic currently being searched, rather than all other users. Our approach yields promising gains in retrieval performance, and has direct implications for improving personalization in search systems.

With more and more people freely express opinions as well as actively interact with each other in discussion threads, online forums are becoming a gold mine with rich information about people's opinions and social behaviors. In this paper, we study an interesting new problem of automatically discovering opposing opinion networks of users from forum discussions, which are subset of users who are strongly against each other on some topic. Toward this goal, we propose to use signals from both textual content (e.g., who says what) and social interactions (e.g., who talks to whom) which are both abundant in online forums. We also design an optimization formulation to combine all the signals in an unsupervised way. We created a data set by manually annotating forum data on five controversial topics and our experimental results show that the proposed optimization method outperforms several baselines and existing approaches, demonstrating the power of combining both text analysis and social network analysis in analyzing and generating the opposing opinion networks.

In contrast to traditional Web search, where topical relevance is often the main selection criterion, news search is characterized by the increased importance of freshness. However, the estimation of relevance and freshness, and especially the relative importance of these two aspects, are highly specific to the query and the time when the query was issued. In this work, we propose a unified framework for modeling the topical relevance and freshness, as well as their relative importance, based on click logs. We use click statistics and content analysis techniques to define a set of temporal features, which predict the right mix of freshness and relevance for a given query. Experimental results on both historical click data and editorial judgments demonstrate the effectiveness of the proposed approach.

Mining detailed opinions buried in the vast amount of review text data is an important, yet quite challenging task with widespread applications in multiple domains. Latent Aspect Rating Analysis (LARA) refers to the task of inferring both opinion ratings on topical aspects (e.g., location, service of a hotel) and the relative weights reviewers have placed on each aspect based on review content and the associated overall ratings. A major limitation of previous work on LARA is the assumption of pre-specified aspects by keywords. However, the aspect information is not always available, and it may be difficult to pre-define appropriate aspects without a good knowledge about what aspects are actually commented on in the reviews.

Online forum discussions are emerging as valuable information repository, where knowledge is accumulated by the interaction among users, leading to multiple threads with structures. Such replying structure in each thread conveys important information about the discussion content. Unfortunately, not all the online forum sites would explicitly record such replying relationship, making it hard to for both users and computers to digest the information buried in a thread discussion.

In this paper, we define and study a new opinionated text data analysis problem called Latent Aspect Rating Analysis (LARA), which aims at analyzing opinions expressed about an entity in an online review at the level of topical aspects to discover each individual reviewer's latent opinion on each aspect as well as the relative emphasis on different aspects when forming the overall judgment of the entity. We propose a novel probabilistic rating regression model to solve this new text mining problem in a general way. Empirical experiments on a hotel review data set show that the proposed latent rating regression model can effectively solve the problem of LARA, and that the detailed analysis of opinions at the level of topical aspects enabled by the proposed model can support a wide range of application tasks, such as aspect opinion summarization, entity ranking based on aspect ratings, and analysis of reviewers rating behavior.

BACKGROUND: Considerable efforts have been made to extract protein-protein interactions from the biological literature, but little work has been done on the extraction of interaction detection methods. It is crucial to annotate the detection methods in the literature, since different detection methods shed different degrees of reliability on the reported interactions. However, the diversity of method mentions in the literature makes the automatic extraction quite challenging. RESULTS: In this article, we develop a generative topic model, the Correlated Method-Word model (CMW model) to extract the detection methods from the literature. In the CMW model, we formulate the correlation between the different methods and related words in a probabilistic framework in order to infer the potential methods from the given document. By applying the model on a corpus of 5319 full text documents annotated by the MINT and IntAct databases, we observe promising results, which outperform the best result reported in the BioCreative II challenge evaluation. CONCLUSION: From the promising experiment results, we can see that the CMW model overcomes the issues caused by the diversity in the method mentions and properly captures the in-depth correlations between the detection methods and related words. The performance outperforming the baseline methods confirms that the dependence assumptions of the model are reasonable and the model is competent for the practical processing.

Traditional discriminative classification method makes little attempt to reveal the probabilistic structure and the correlation within both input and output spaces. In the scenario of multi-label classification, most of the classifiers simply assume the predefined classes are independently distributed, which would definitely hinder the classification performance when there are intrinsic correlations between the classes. In this article, we propose a generative probabilistic model, the Correlated Labeling Model (CoL Model), to formulate the correlation between different classes. The CoL model is presented to capture the correlation between classes and the underlying structures via the latent random variables in a supervised manner. We develop a variational procedure to approximate the posterior distribution and employ the EM algorithm for the empirical Bayes parameter estimation. In our evaluations, the proposed model achieved promising results on various data sets.

BACKGROUND: Efficient features play an important role in automated text classification, which definitely facilitates the access of large-scale data. In the bioscience field, biological structures and terminologies are described by a large number of features; domain dependent features would significantly improve the classification performance. How to effectively select and integrate different types of features to improve the biological literature classification performance is the major issue studied in this paper. RESULTS: To efficiently classify the biological literatures, we propose a novel feature value schema TF*ML, features covering from lower level domain independent "string feature" to higher level domain dependent "semantic template feature", and proper integrations among the features. Compared to our previous approaches, the performance is improved in terms of AUC and F-Score by 11.5% and 8.8% respectively, and outperforms the best performance achieved in BioCreAtIvE 2006. CONCLUSIONS: Different types of features possess different discriminative capabilities in literature classification; proper integration of domain independent and dependent features would significantly improve the performance and overcome the over-fitting on data distribution.