Fuzheng Zhang

Yuchong Sun, Che Liu, Kun Zhou et al.

Humans often interact with large language models (LLMs) in multi-turn interaction to obtain desired answers or more information. However, most existing studies overlook the multi-turn instruction following ability of LLMs, in terms of training dataset, training method, and evaluation benchmark. In this paper, we introduce Parrot, a solution aiming to enhance multi-turn instruction following for LLMs. First, we introduce an efficient but effective method for collecting multi-turn instructions that feature human-like queries, such as anaphora and ellipsis. Second, we propose a context-aware preference optimization strategy to further enhance LLMs for complex queries in multi-turn interaction. Moreover, to quantitatively evaluate LLMs in multi-turn instruction following, we manually build a multi-turn benchmark derived from existing ones. Extensive experiments show that Parrot improves current LLMs by up to 7.2% in multi-turn instruction following. Our dataset and codes will be open-sourced to facilitate future research.

Xing Wu, Guangyuan Ma, Peng Wang et al.

Growing techniques have been emerging to improve the performance of passage retrieval. As an effective representation bottleneck pretraining technique, the contextual masked auto-encoder utilizes contextual embedding to assist in the reconstruction of passages. However, it only uses a single auto-encoding pre-task for dense representation pre-training. This study brings multi-view modeling to the contextual masked auto-encoder. Firstly, multi-view representation utilizes both dense and sparse vectors as multi-view representations, aiming to capture sentence semantics from different aspects. Moreover, multiview decoding paradigm utilizes both autoencoding and auto-regressive decoders in representation bottleneck pre-training, aiming to provide both reconstructive and generative signals for better contextual representation pretraining. We refer to this multi-view pretraining method as CoT-MAE v2. Through extensive experiments, we show that CoT-MAE v2 is effective and robust on large-scale passage retrieval benchmarks and out-of-domain zero-shot benchmarks.

Jiao Ou, Junda Lu, Che Liu et al.

Large language models (LLMs) have achieved remarkable breakthroughs in new dialogue capabilities by leveraging instruction tuning, which refreshes human impressions of dialogue systems. The long-standing goal of dialogue systems is to be human-like enough to establish long-term connections with users. Therefore, there has been an urgent need to evaluate LLMs as human-like dialogue systems. In this paper, we propose DialogBench, a dialogue evaluation benchmark that contains 12 dialogue tasks to probe the capabilities of LLMs as human-like dialogue systems should have. Specifically, we prompt GPT-4 to generate evaluation instances for each task. We first design the basic prompt based on widely used design principles and further mitigate the existing biases to generate higher-quality evaluation instances. Our extensive tests on English and Chinese DialogBench of 26 LLMs show that instruction tuning improves the human likeness of LLMs to a certain extent, but most LLMs still have much room for improvement as human-like dialogue systems. Interestingly, results also show that the positioning of assistant AI can make instruction tuning weaken the human emotional perception of LLMs and their mastery of information about human daily life.

Lei Lin, Jiayi Fu, Pengli Liu et al.

Although chain-of-thought (CoT) prompting combined with language models has achieved encouraging results on complex reasoning tasks, the naive greedy decoding used in CoT prompting usually causes the repetitiveness and local optimality. To address this shortcoming, ensemble-optimization tries to obtain multiple reasoning paths to get the final answer assembly. However, current ensemble-optimization methods either simply employ rule-based post-processing such as \textit{self-consistency}, or train an additional model based on several task-related human annotations to select the best one among multiple reasoning paths, yet fail to generalize to realistic settings where the type of input questions is unknown or the answer format of reasoning paths is unknown. To avoid their limitations, we propose \textbf{Self-Agreement}, a generalizable ensemble-optimization method applying in almost all scenarios where the type of input questions and the answer format of reasoning paths may be known or unknown. Self-agreement firstly samples from language model's decoder to generate a \textit{diverse} set of reasoning paths, and subsequently prompts the language model \textit{one more time} to determine the optimal answer by selecting the most \textit{agreed} answer among the sampled reasoning paths. Self-agreement simultaneously achieves remarkable performance on six public reasoning benchmarks and superior generalization capabilities.

Yihong Tang, Jiao Ou, Che Liu et al.

Role-playing is an emerging application in the field of Human-Computer Interaction (HCI), primarily implemented through the alignment training of a large language model (LLM) with assigned characters. Despite significant progress, role-playing agents (RPLAs) still struggle with maintaining role-consistency across conversations, particularly when confronted with boundary queries subtly related to character attributes. In this paper, we present ERABAL, a framework aimed at enhancing RPLAs' role-playing capabilities through boundary-aware learning. ERABAL encompasses a generation pipeline for role-specific dialogues and a concomitant methodology for alignment training. Through comprehensive evaluations, we demonstrate that ERABAL is both efficient and effective. By training with significantly fewer dialogues than those used in leading approaches, ERABAL achieves notable improvements across WikiRoleEval, CharacterEval, and the role-playing subset of MT-Bench compared to the generalist baseline models. Our code and datasets will be made publicly available to support further research.

Jiayi Fu, Lei Lin, Xiaoyang Gao et al.

Recent advancements in large language models (LLMs) have demonstrated remarkable abilities in handling a variety of natural language processing (NLP) downstream tasks, even on mathematical tasks requiring multi-step reasoning. In this report, we introduce the KwaiYiiMath which enhances the mathematical reasoning abilities of KwaiYiiBase1, by applying Supervised Fine-Tuning (SFT) and Reinforced Learning from Human Feedback (RLHF), including on both English and Chinese mathematical tasks. Meanwhile, we also constructed a small-scale Chinese primary school mathematics test set (named KMath), consisting of 188 examples to evaluate the correctness of the problem-solving process generated by the models. Empirical studies demonstrate that KwaiYiiMath can achieve state-of-the-art (SOTA) performance on GSM8k, CMath, and KMath compared with the similar size models, respectively.

Zhipeng Chen, Kun Zhou, Wayne Xin Zhao et al.

Reinforcement learning (RL) has been widely used in training large language models (LLMs) for preventing unexpected outputs, eg reducing harmfulness and errors. However, existing RL methods mostly adopt the instance-level reward, which is unable to provide fine-grained supervision for complex reasoning tasks, and can not focus on the few key tokens that lead to the incorrectness. To address it, we propose a new RL method named RLMEC that incorporates a generative model as the reward model, which is trained by the erroneous solution rewriting task under the minimum editing constraint, and can produce token-level rewards for RL training. Based on the generative reward model, we design the token-level RL objective for training and an imitation-based regularization for stabilizing RL process. And the both objectives focus on the learning of the key tokens for the erroneous solution, reducing the effect of other unimportant tokens. The experiment results on mathematical tasks and question-answering tasks have demonstrated the effectiveness of our approach. Our code and data are available at https://github.com/RUCAIBox/RLMEC.

Xiao Wang, Jianlong Wu, Zijia Lin et al.

Recently, video-language understanding has achieved great success through large-scale pre-training. However, data scarcity remains a prevailing challenge. This study quantitatively reveals an "impossible trinity" among data quantity, diversity, and quality in pre-training datasets. Recent efforts seek to refine large-scale, diverse ASR datasets compromised by low quality through synthetic annotations. These methods successfully leverage useful information in multimodal video content (frames, tags, ASR transcripts, etc.) to refine the original annotations. Nevertheless, they struggle to mitigate noise within synthetic annotations and lack scalability as the dataset size expands. To address these issues, we introduce the Video DataFlywheel framework, which iteratively refines video annotations with improved noise control methods. For iterative refinement, we first leverage a video-language model to generate synthetic annotations, resulting in a refined dataset. Then, we pre-train on it and fine-tune on human refinement examples for a stronger model. These processes are repeated for continuous improvement. For noise control, we present AdaTaiLr, a novel noise control method that requires weaker assumptions on noise distribution, thereby proving more effective in large datasets with theoretical guarantees. The combination of iterative refinement and AdaTaiLr can achieve better scalability in video-language understanding. Extensive experiments show that our framework outperforms existing data refinement baselines, delivering a 3% performance boost and improving dataset quality with minimal diversity loss. Furthermore, our refined dataset facilitates significant improvements in various video-language understanding tasks, including video question answering and text-video retrieval.

Jiakang Wang, Runze Liu, Fuzheng Zhang et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has become an effective post-training method for improving the reasoning abilities of Large Language Models (LLMs), mainly by shaping higher-order behaviors such as reflection and planning. However, previous RLVR algorithms often apply uniform training signals to all tokens, without considering the different roles of low-entropy knowledge-related tokens and high-entropy reasoning-related tokens. Some recent methods try to separate these token types by gradient masking or asynchronous updates, but these approaches may break semantic dependencies in the model output and hinder effective learning. In this work, we propose Archer, an entropy-aware RLVR approach with dual-token constraints and synchronous updates. Specifically, our method applies weaker KL regularization and higher clipping thresholds to reasoning tokens to encourage exploration, while using stronger constraints on knowledge tokens to maintain factual knowledge. Experimental results on several mathematical reasoning and code generation benchmarks show that our approach significantly outperforms previous RLVR methods, reaching or exceeding state-of-the-art performance among models of comparable size. The code is available at https://github.com/wizard-III/ArcherCodeR.

Xinlong Chen, Yuanxing Zhang, Chongling Rao et al.

The training of controllable text-to-video (T2V) models relies heavily on the alignment between videos and captions, yet little existing research connects video caption evaluation with T2V generation assessment. This paper introduces VidCapBench, a video caption evaluation scheme specifically designed for T2V generation, agnostic to any particular caption format. VidCapBench employs a data annotation pipeline, combining expert model labeling and human refinement, to associate each collected video with key information spanning video aesthetics, content, motion, and physical laws. VidCapBench then partitions these key information attributes into automatically assessable and manually assessable subsets, catering to both the rapid evaluation needs of agile development and the accuracy requirements of thorough validation. By evaluating numerous state-of-the-art captioning models, we demonstrate the superior stability and comprehensiveness of VidCapBench compared to existing video captioning evaluation approaches. Verification with off-the-shelf T2V models reveals a significant positive correlation between scores on VidCapBench and the T2V quality evaluation metrics, indicating that VidCapBench can provide valuable guidance for training T2V models. The project is available at https://github.com/VidCapBench/VidCapBench.

Hongzhi Zhang, Jia Fu, Jingyuan Zhang et al.

Reinforcement learning (RL) for large language models is an energy-intensive endeavor: training can be unstable, and the policy may gradually drift away from its pretrained weights. We present \emph{RLEP}\, -- \,Reinforcement Learning with Experience rePlay\, -- \,a two-phase framework that first collects verified trajectories and then replays them during subsequent training. At every update step, the policy is optimized on mini-batches that blend newly generated rollouts with these replayed successes. By replaying high-quality examples, RLEP steers the model away from fruitless exploration, focuses learning on promising reasoning paths, and delivers both faster convergence and stronger final performance. On the Qwen2.5-Math-7B base model, RLEP reaches baseline peak accuracy with substantially fewer updates and ultimately surpasses it, improving accuracy on AIME-2024 from 38.2% to 39.9%, on AIME-2025 from 19.8% to 22.3%, and on AMC-2023 from 77.0% to 82.2%. Our code, datasets, and checkpoints are publicly available at https://github.com/Kwai-Klear/RLEP to facilitate reproducibility and further research.

Yulan Hu, Sheng Ouyang, Jingyu Liu et al.

Graph contrastive learning (GCL) has emerged as a representative graph self-supervised method, achieving significant success. The currently prevalent optimization objective for GCL is InfoNCE. Typically, it employs augmentation techniques to obtain two views, where a node in one view acts as the anchor, the corresponding node in the other view serves as the positive sample, and all other nodes are regarded as negative samples. The goal is to minimize the distance between the anchor node and positive samples and maximize the distance to negative samples. However, due to the lack of label information during training, InfoNCE inevitably treats samples from the same class as negative samples, leading to the issue of false negative samples. This can impair the learned node representations and subsequently hinder performance in downstream tasks. While numerous methods have been proposed to mitigate the impact of false negatives, they still face various challenges. For instance, while increasing the number of negative samples can dilute the impact of false negatives, it concurrently increases computational burden. Thus, we propose GraphRank, a simple yet efficient graph contrastive learning method that addresses the problem of false negative samples by redefining the concept of negative samples to a certain extent, thereby avoiding the issue of false negative samples. The effectiveness of GraphRank is empirically validated through experiments on the node, edge, and graph level tasks.

Hao Yi, Qingyang Li, Yulan Hu et al.

Recently, enhancing the numerical and logical reasoning capability of Large Language Models (LLMs) has emerged as a research hotspot. Existing methods face several limitations: inference-phase techniques (e.g., Chain of Thoughts) rely on prompt selection and the pretrained knowledge; sentence-level Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) struggle with step-wise mathematical correctness and depend on stronger models distillation or human annotations; while Reinforcement Learning (RL) approaches incur high GPU memory costs and unstable training. To address these, we propose \textbf{S}elf-training framework integrating \textbf{P}rocess \textbf{P}reference learning using \textbf{D}ynamic value margin (SPPD). SPPD leverages a process-based Markov Decision Process (MDP) and Bellman optimality equation to derive \textbf{dynamic value margin} on step-level preference optimization, which employs tree-based self-sampling on model responses \textbf{without any distillation} from other models. Furthermore, we theoretically prove that SPPD is \textbf{equivalent to on-policy policy gradient methods} under reward constraints. Experiments on 7B-scale models demonstrate superior performance across in-domain and out-domain mathematical benchmarks. We open-source our code at \href{https://anonymous.4open.science/r/SSDPO-D-DCDD}{https://anonymous.4open.science/r/SPPD-DCDD}.

Jingyuan Zhang, Hongzhi Zhang, Zhou Haonan et al.

Data curation plays a crucial role in training powerful Visual Language Models (VLMs). In this work, we introduce the concept of Data Metabolism and present our data-centric framework to build VLMs throughout the development lifecycle. Starting from a standard model architecture, we discuss and provide insights into two crucial development steps: data curation and iteration, forming a closed-loop system that continuously improves model performance. We show a detailed codebook on how to process existing massive datasets and build user-specific data flywheel. As a demonstration, we release a VLM, named Capybara-VL, which excels in typical multimodal tasks (e.g. , visual question answering, scientific reasoning, and text-rich tasks). Despite its relatively compact size, Capybara-VL surpasses several open-source models that are up to 10 times larger in size. Moreover, it achieves results that are on par with those of several leading proprietary models, demonstrating its remarkable competitiveness. These results highlight the power of our data-centric framework and the potential of training smaller and more efficient VLMs.

Jiakang Wang, Runze Liu, Lei Lin et al.

Recent Large Language Model (LLM) post-training methods rely on token-level clipping mechanisms during Reinforcement Learning (RL). However, we identify a fundamental flaw in this Outcome-Supervised RL (OSRL) paradigm: the Importance Sampling (IS) ratios of positive-advantage tokens are mismatched, leading to unbalanced token weighting for positive and negative tokens. This mismatch suppresses the update of low-probability tokens while over-amplifying already high-probability ones. To address this, we propose Asymmetric Importance Sampling Policy Optimization (ASPO), which uses a simple yet effective strategy that flips the IS ratios of positive-advantage tokens, aligning their update direction with the learning dynamics of negative ones. AIS further incorporates a soft dual-clipping mechanism to stabilize extreme updates while maintaining gradient flow. Comprehensive experiments on coding and mathematical reasoning benchmarks demonstrate that ASPO significantly mitigates premature convergence, improves training stability, and enhances final performance over strong GRPO-based baselines. Our analysis provides new insights into the role of token-level weighting in OSRL and highlights the critical importance of correcting IS in LLM RL. The code and models of ASPO are available at https://github.com/wizard-III/Archer2.0.

Kun Zhou, Wayne Xin Zhao, Sirui Wang et al.

Recent works have shown that powerful pre-trained language models (PLM) can be fooled by small perturbations or intentional attacks. To solve this issue, various data augmentation techniques are proposed to improve the robustness of PLMs. However, it is still challenging to augment semantically relevant examples with sufficient diversity. In this work, we present Virtual Data Augmentation (VDA), a general framework for robustly fine-tuning PLMs. Based on the original token embeddings, we construct a multinomial mixture for augmenting virtual data embeddings, where a masked language model guarantees the semantic relevance and the Gaussian noise provides the augmentation diversity. Furthermore, a regularized training strategy is proposed to balance the two aspects. Extensive experiments on six datasets show that our approach is able to improve the robustness of PLMs and alleviate the performance degradation under adversarial attacks. Our codes and data are publicly available at \textcolor{blue}{\url{https://github.com/RUCAIBox/VDA}}.

Yihong Tang, Jiao Ou, Che Liu et al.

The advent of Large Language Models (LLMs) has propelled dialogue generation into new realms, particularly in the field of role-playing systems (RPSs). While enhanced with ordinary role-relevant training dialogues, existing LLM-based RPSs still struggle to align with roles when handling intricate and trapped queries in boundary scenarios. In this paper, we design the Modular ORchestrated Trap-setting Interaction SystEm (MORTISE) to benchmark and improve the role-playing LLMs' performance. MORTISE can produce highly role-relevant aggressive queries through the collaborative effort of multiple LLM-based modules, and formulate corresponding responses to create an adversarial training dataset via a consistent response generator. We select 190 Chinese and English roles to construct aggressive queries to benchmark existing role-playing LLMs. Through comprehensive evaluation, we find that existing models exhibit a general deficiency in role alignment capabilities. We further select 180 of the roles to collect an adversarial training dataset (named RoleAD) and retain the other 10 roles for testing. Experiments on models improved by RoleAD indicate that our adversarial dataset ameliorates this deficiency, with the improvements demonstrating a degree of generalizability in ordinary scenarios.

Sheng Ouyang, Yulan Hu, Ge Chen et al.

Rewards serve as proxies for human preferences and play a crucial role in Reinforcement Learning from Human Feedback (RLHF). However, if these rewards are inherently imperfect, exhibiting various biases, they can adversely affect the alignment of large language models (LLMs). In this paper, we collectively define the various biases present in rewards as the problem of reward unfairness. We propose a bias-agnostic method to address the issue of reward fairness from a resource allocation perspective, without specifically designing for each type of bias, yet effectively mitigating them. Specifically, we model preference learning as a resource allocation problem, treating rewards as resources to be allocated while considering the trade-off between utility and fairness in their distribution. We propose two methods, Fairness Regularization and Fairness Coefficient, to achieve fairness in rewards. We apply our methods in both verification and reinforcement learning scenarios to obtain a fairness reward model and a policy model, respectively. Experiments conducted in these scenarios demonstrate that our approach aligns LLMs with human preferences in a more fair manner.

Lingyu Kong, Hongzhi Zhang, Jingyuan Zhang et al.

Current vision-language models (VLMs) have demonstrated remarkable capabilities across diverse video understanding applications. Designing VLMs for video inputs requires effectively modeling the temporal dimension (i.e. capturing dependencies across frames) and balancing the processing of short and long videos. Specifically, short videos demand preservation of fine-grained details, whereas long videos require strategic compression of visual information to handle extensive temporal contexts efficiently. However, our empirical analysis reveals a critical limitation: most existing VLMs suffer severe performance degradation in long video understanding tasks when compressing visual tokens below a quarter of their original visual tokens. To enable more effective modeling of both short and long video inputs, we propose Clapper, a method that utilizes a slow-fast strategy for video representation and introduces a novel module named TimePerceiver for efficient temporal-spatial encoding within existing VLM backbones. By using our method, we achieves 13x compression of visual tokens per frame (averaging 61 tokens/frame) without compromising QA accuracy. In our experiments, Clapper achieves 62.0% on VideoMME, 69.8% on MLVU, and 67.4% on TempCompass, all with fewer than 6,000 visual tokens per video. The code will be publicly available on the homepage.

Hao Yi, Qingyang Li, Yulan Hu et al.

High-quality video-text preference data is crucial for Multimodal Large Language Models (MLLMs) alignment. However, existing preference data is very scarce. Obtaining VQA preference data for preference training is costly, and manually annotating responses is highly unreliable, which could result in low-quality pairs. Meanwhile, AI-generated responses controlled by temperature adjustment lack diversity. To address these issues, we propose a high-quality VQA preference dataset, called \textit{\textbf{M}ultiple \textbf{M}ultimodal \textbf{A}rtificial \textbf{I}ntelligence \textbf{P}reference Datasets in \textbf{V}QA} (\textbf{MMAIP-V}), which is constructed by sampling from the response distribution set and using an external scoring function for response evaluation. Furthermore, to fully leverage the preference knowledge in MMAIP-V and ensure sufficient optimization, we propose \textit{\textbf{Iter}ative \textbf{W}eak-to-\textbf{S}trong \textbf{R}einforcement \textbf{L}earning from \textbf{AI} \textbf{F}eedback for video MLLMs} (\textbf{Iter-W2S-RLAIF}), a framework that gradually enhances MLLMs' alignment capabilities by iteratively updating the reference model and performing parameter extrapolation. Finally, we propose an unbiased and information-complete evaluation scheme in VQA evaluation. Experiments demonstrate that MMAIP-V is beneficial for MLLMs in preference learning and Iter-W2S-RLAIF fully exploits the alignment information in MMAIP-V. We believe that the proposed automatic VQA preference data generation pipeline based on AI feedback can greatly promote future work in the MLLMs alignment. \textbf{Code and dataset are available} \href{https://anonymous.4open.science/r/MMAIP-V_Iter-W2S-RLAIF-702F}{MMAIP-V\_Iter-W2S-RLAIF-702F}.

Yulan Hu, Qingyang Li, Sheng Ouyang et al.

Reinforcement Learning from Human Feedback (RLHF) facilitates the alignment of large language models (LLMs) with human preferences, thereby enhancing the quality of responses generated. A critical component of RLHF is the reward model, which is trained on preference data and outputs a scalar reward during the inference stage. However, the collection of preference data still lacks thorough investigation. Recent studies indicate that preference data is collected either by AI or humans, where chosen and rejected instances are identified among pairwise responses. We question whether this process effectively filters out noise and ensures sufficient diversity in collected data. To address these concerns, for the first time, we propose a comprehensive framework for preference data collection, decomposing the process into four incremental steps: Prompt Generation, Response Generation, Response Filtering, and Human Labeling. This structured approach ensures the collection of high-quality preferences while reducing reliance on human labor. We conducted comprehensive experiments based on the data collected at different stages, demonstrating the effectiveness of the proposed data collection method.

Kaiwen Wei, Jingyuan Zhang, Hongzhi Zhang et al.

Large Language Models (LLMs) exhibit remarkable generative capabilities, enabling the generation of valuable information. Despite these advancements, previous research found that LLMs sometimes struggle with adhering to specific constraints (e.g., in specific place or at specific time), at times even overlooking them, which leads to responses that are either too generic or not fully satisfactory. Existing approaches attempted to address this issue by decomposing or rewriting input instructions, yet they fall short in adequately emphasizing specific constraints and in unlocking the underlying knowledge (e.g., programming within the context of software development). In response, this paper proposes a simple yet effective method named Chain-of-Specificity (CoS). Specifically, CoS iteratively emphasizes the specific constraints in the input instructions, unlocks knowledge within LLMs, and refines responses. Experiments conducted on publicly available and self-build complex datasets demonstrate that CoS outperforms existing methods in enhancing generated content especially for the specificity. Besides, as the number of specific constraints increase, other baselines falter, while CoS still performs well. Moreover, we show that distilling responses generated by CoS effectively enhances the ability of smaller models to follow the constrained instructions. Resources of this paper will be released for further research.

Yile Chen, Xiucheng Li, Gao Cong et al.

As a fundamental component in location-based services, inferring the relationship between points-of-interests (POIs) is very critical for service providers to offer good user experience to business owners and customers. Most of the existing methods for relationship inference are not targeted at POI, thus failing to capture unique spatial characteristics that have huge effects on POI relationships. In this work we propose PRIM to tackle POI relationship inference for multiple relation types. PRIM features four novel components, including a weighted relational graph neural network, category taxonomy integration, a self-attentive spatial context extractor, and a distance-specific scoring function. Extensive experiments on two real-world datasets show that PRIM achieves the best results compared to state-of-the-art baselines and it is robust against data sparsity and is applicable to unseen cases in practice.

Zihao Zhao, Jiawei Chen, Sheng Zhou et al.

Recommender system usually suffers from severe popularity bias -- the collected interaction data usually exhibits quite imbalanced or even long-tailed distribution over items. Such skewed distribution may result from the users' conformity to the group, which deviates from reflecting users' true preference. Existing efforts for tackling this issue mainly focus on completely eliminating popularity bias. However, we argue that not all popularity bias is evil. Popularity bias not only results from conformity but also item quality, which is usually ignored by existing methods. Some items exhibit higher popularity as they have intrinsic better property. Blindly removing the popularity bias would lose such important signal, and further deteriorate model performance. To sufficiently exploit such important information for recommendation, it is essential to disentangle the benign popularity bias caused by item quality from the harmful popularity bias caused by conformity. Although important, it is quite challenging as we lack an explicit signal to differentiate the two factors of popularity bias. In this paper, we propose to leverage temporal information as the two factors exhibit quite different patterns along the time: item quality revealing item inherent property is stable and static while conformity that depends on items' recent clicks is highly time-sensitive. Correspondingly, we further propose a novel Time-aware DisEntangled framework (TIDE), where a click is generated from three components namely the static item quality, the dynamic conformity effect, as well as the user-item matching score returned by any recommendation model. Lastly, we conduct interventional inference such that the recommendation can benefit from the benign popularity bias while circumvent the harmful one. Extensive experiments on three real-world datasets demonstrated the effectiveness of TIDE.

Yuanmeng Yan, Rumei Li, Sirui Wang et al.

Learning high-quality sentence representations benefits a wide range of natural language processing tasks. Though BERT-based pre-trained language models achieve high performance on many downstream tasks, the native derived sentence representations are proved to be collapsed and thus produce a poor performance on the semantic textual similarity (STS) tasks. In this paper, we present ConSERT, a Contrastive Framework for Self-Supervised Sentence Representation Transfer, that adopts contrastive learning to fine-tune BERT in an unsupervised and effective way. By making use of unlabeled texts, ConSERT solves the collapse issue of BERT-derived sentence representations and make them more applicable for downstream tasks. Experiments on STS datasets demonstrate that ConSERT achieves an 8\% relative improvement over the previous state-of-the-art, even comparable to the supervised SBERT-NLI. And when further incorporating NLI supervision, we achieve new state-of-the-art performance on STS tasks. Moreover, ConSERT obtains comparable results with only 1000 samples available, showing its robustness in data scarcity scenarios.

Hongyin Tang, Xingwu Sun, Beihong Jin et al.

Recently, the retrieval models based on dense representations have been gradually applied in the first stage of the document retrieval tasks, showing better performance than traditional sparse vector space models. To obtain high efficiency, the basic structure of these models is Bi-encoder in most cases. However, this simple structure may cause serious information loss during the encoding of documents since the queries are agnostic. To address this problem, we design a method to mimic the queries on each of the documents by an iterative clustering process and represent the documents by multiple pseudo queries (i.e., the cluster centroids). To boost the retrieval process using approximate nearest neighbor search library, we also optimize the matching function with a two-step score calculation procedure. Experimental results on several popular ranking and QA datasets show that our model can achieve state-of-the-art results.

Jiahao Bu, Lei Ren, Shuang Zheng et al.

Sentiment analysis has attracted increasing attention in e-commerce. The sentiment polarities underlying user reviews are of great value for business intelligence. Aspect category sentiment analysis (ACSA) and review rating prediction (RP) are two essential tasks to detect the fine-to-coarse sentiment polarities. %Considering the sentiment of the aspects(ACSA) and the overall review rating(RP) simultaneously has the potential to improve the overall performance. ACSA and RP are highly correlated and usually employed jointly in real-world e-commerce scenarios. While most public datasets are constructed for ACSA and RP separately, which may limit the further exploitation of both tasks. To address the problem and advance related researches, we present a large-scale Chinese restaurant review dataset \textbf{ASAP} including $46,730$ genuine reviews from a leading online-to-offline (O2O) e-commerce platform in China. Besides a $5$-star scale rating, each review is manually annotated according to its sentiment polarities towards $18$ pre-defined aspect categories. We hope the release of the dataset could shed some light on the fields of sentiment analysis. Moreover, we propose an intuitive yet effective joint model for ACSA and RP. Experimental results demonstrate that the joint model outperforms state-of-the-art baselines on both tasks.

Yang Yang, Junmei Hao, Canjia Li et al.

As a concise form of user reviews, tips have unique advantages to explain the search results, assist users' decision making, and further improve user experience in vertical search scenarios. Existing work on tip generation does not take query into consideration, which limits the impact of tips in search scenarios. To address this issue, this paper proposes a query-aware tip generation framework, integrating query information into encoding and subsequent decoding processes. Two specific adaptations of Transformer and Recurrent Neural Network (RNN) are proposed. For Transformer, the query impact is incorporated into the self-attention computation of both the encoder and the decoder. As for RNN, the query-aware encoder adopts a selective network to distill query-relevant information from the review, while the query-aware decoder integrates the query information into the attention computation during decoding. The framework consistently outperforms the competing methods on both public and real-world industrial datasets. Last but not least, online deployment experiments on Dianping demonstrate the advantage of the proposed framework for tip generation as well as its online business values.

Yu Yin, Zhenya Huang, Enhong Chen et al.

Transcribing content from structural images, e.g., writing notes from music scores, is a challenging task as not only the content objects should be recognized, but the internal structure should also be preserved. Existing image recognition methods mainly work on images with simple content (e.g., text lines with characters), but are not capable to identify ones with more complex content (e.g., structured symbols), which often follow a fine-grained grammar. To this end, in this paper, we propose a hierarchical Spotlight Transcribing Network (STN) framework followed by a two-stage "where-to-what" solution. Specifically, we first decide "where-to-look" through a novel spotlight mechanism to focus on different areas of the original image following its structure. Then, we decide "what-to-write" by developing a GRU based network with the spotlight areas for transcribing the content accordingly. Moreover, we propose two implementations on the basis of STN, i.e., STNM and STNR, where the spotlight movement follows the Markov property and Recurrent modeling, respectively. We also design a reinforcement method to refine the framework by self-improving the spotlight mechanism. We conduct extensive experiments on many structural image datasets, where the results clearly demonstrate the effectiveness of STN framework.

Bowen Xing, Lejian Liao, Dandan Song et al.

Aspect-based sentiment analysis (ABSA) aims to predict fine-grained sentiments of comments with respect to given aspect terms or categories. In previous ABSA methods, the importance of aspect has been realized and verified. Most existing LSTM-based models take aspect into account via the attention mechanism, where the attention weights are calculated after the context is modeled in the form of contextual vectors. However, aspect-related information may be already discarded and aspect-irrelevant information may be retained in classic LSTM cells in the context modeling process, which can be improved to generate more effective context representations. This paper proposes a novel variant of LSTM, termed as aspect-aware LSTM (AA-LSTM), which incorporates aspect information into LSTM cells in the context modeling stage before the attention mechanism. Therefore, our AA-LSTM can dynamically produce aspect-aware contextual representations. We experiment with several representative LSTM-based models by replacing the classic LSTM cells with the AA-LSTM cells. Experimental results on SemEval-2014 Datasets demonstrate the effectiveness of AA-LSTM.

Hongwei Wang, Fuzheng Zhang, Mengdi Zhang et al.

Knowledge graphs capture structured information and relations between a set of entities or items. As such knowledge graphs represent an attractive source of information that could help improve recommender systems. However, existing approaches in this domain rely on manual feature engineering and do not allow for an end-to-end training. Here we propose Knowledge-aware Graph Neural Networks with Label Smoothness regularization (KGNN-LS) to provide better recommendations. Conceptually, our approach computes user-specific item embeddings by first applying a trainable function that identifies important knowledge graph relationships for a given user. This way we transform the knowledge graph into a user-specific weighted graph and then apply a graph neural network to compute personalized item embeddings. To provide better inductive bias, we rely on label smoothness assumption, which posits that adjacent items in the knowledge graph are likely to have similar user relevance labels/scores. Label smoothness provides regularization over the edge weights and we prove that it is equivalent to a label propagation scheme on a graph. We also develop an efficient implementation that shows strong scalability with respect to the knowledge graph size. Experiments on four datasets show that our method outperforms state of the art baselines. KGNN-LS also achieves strong performance in cold-start scenarios where user-item interactions are sparse.

Hongwei Wang, Fuzheng Zhang, Miao Zhao et al.

Collaborative filtering often suffers from sparsity and cold start problems in real recommendation scenarios, therefore, researchers and engineers usually use side information to address the issues and improve the performance of recommender systems. In this paper, we consider knowledge graphs as the source of side information. We propose MKR, a Multi-task feature learning approach for Knowledge graph enhanced Recommendation. MKR is a deep end-to-end framework that utilizes knowledge graph embedding task to assist recommendation task. The two tasks are associated by cross&compress units, which automatically share latent features and learn high-order interactions between items in recommender systems and entities in the knowledge graph. We prove that cross&compress units have sufficient capability of polynomial approximation, and show that MKR is a generalized framework over several representative methods of recommender systems and multi-task learning. Through extensive experiments on real-world datasets, we demonstrate that MKR achieves substantial gains in movie, book, music, and news recommendation, over state-of-the-art baselines. MKR is also shown to be able to maintain a decent performance even if user-item interactions are sparse.

Fuzhen Zhuang, Yingmin Zhou, Fuzheng Zhang et al.

Transfer learning has attracted a large amount of interest and research in last decades, and some efforts have been made to build more precise recommendation systems. Most previous transfer recommendation systems assume that the target domain shares the same/similar rating patterns with the auxiliary source domain, which is used to improve the recommendation performance. However, to the best of our knowledge, almost these works do not consider the characteristics of sequential data. In this paper, we study the new cross-domain recommendation scenario for mining novelty-seeking trait. Recent studies in psychology suggest that novelty-seeking trait is highly related to consumer behavior, which has a profound business impact on online recommendation. Previous work performing on only one single target domain may not fully characterize users' novelty-seeking trait well due to the data scarcity and sparsity, leading to the poor recommendation performance. Along this line, we proposed a new cross-domain novelty-seeking trait mining model (CDNST for short) to improve the sequential recommendation performance by transferring the knowledge from auxiliary source domain. We conduct systematic experiments on three domain data sets crawled from Douban (www.douban.com) to demonstrate the effectiveness of the proposed model. Moreover, we analyze how the temporal property of sequential data affects the performance of CDNST, and conduct simulation experiments to validate our analysis.

Hongwei Wang, Fuzheng Zhang, Xing Xie et al.

Online news recommender systems aim to address the information explosion of news and make personalized recommendation for users. In general, news language is highly condensed, full of knowledge entities and common sense. However, existing methods are unaware of such external knowledge and cannot fully discover latent knowledge-level connections among news. The recommended results for a user are consequently limited to simple patterns and cannot be extended reasonably. Moreover, news recommendation also faces the challenges of high time-sensitivity of news and dynamic diversity of users' interests. To solve the above problems, in this paper, we propose a deep knowledge-aware network (DKN) that incorporates knowledge graph representation into news recommendation. DKN is a content-based deep recommendation framework for click-through rate prediction. The key component of DKN is a multi-channel and word-entity-aligned knowledge-aware convolutional neural network (KCNN) that fuses semantic-level and knowledge-level representations of news. KCNN treats words and entities as multiple channels, and explicitly keeps their alignment relationship during convolution. In addition, to address users' diverse interests, we also design an attention module in DKN to dynamically aggregate a user's history with respect to current candidate news. Through extensive experiments on a real online news platform, we demonstrate that DKN achieves substantial gains over state-of-the-art deep recommendation models. We also validate the efficacy of the usage of knowledge in DKN.

Hongwei Wang, Jia Wang, Jialin Wang et al.

The goal of graph representation learning is to embed each vertex in a graph into a low-dimensional vector space. Existing graph representation learning methods can be classified into two categories: generative models that learn the underlying connectivity distribution in the graph, and discriminative models that predict the probability of edge existence between a pair of vertices. In this paper, we propose GraphGAN, an innovative graph representation learning framework unifying above two classes of methods, in which the generative model and discriminative model play a game-theoretical minimax game. Specifically, for a given vertex, the generative model tries to fit its underlying true connectivity distribution over all other vertices and produces "fake" samples to fool the discriminative model, while the discriminative model tries to detect whether the sampled vertex is from ground truth or generated by the generative model. With the competition between these two models, both of them can alternately and iteratively boost their performance. Moreover, when considering the implementation of generative model, we propose a novel graph softmax to overcome the limitations of traditional softmax function, which can be proven satisfying desirable properties of normalization, graph structure awareness, and computational efficiency. Through extensive experiments on real-world datasets, we demonstrate that GraphGAN achieves substantial gains in a variety of applications, including link prediction, node classification, and recommendation, over state-of-the-art baselines.