Bing Qin

Zheng Chu, Jingchang Chen, Qianglong Chen et al.

Reasoning, a fundamental cognitive process integral to human intelligence, has garnered substantial interest within artificial intelligence. Notably, recent studies have revealed that chain-of-thought prompting significantly enhances LLM's reasoning capabilities, which attracts widespread attention from both academics and industry. In this paper, we systematically investigate relevant research, summarizing advanced methods through a meticulous taxonomy that offers novel perspectives. Moreover, we delve into the current frontiers and delineate the challenges and future directions, thereby shedding light on future research. Furthermore, we engage in a discussion about open questions. We hope this paper serves as an introduction for beginners and fosters future research. Resources have been made publicly available at https://github.com/zchuz/CoT-Reasoning-Survey

Weixiang Zhao, Yanyan Zhao, Zhuojun Li et al.

Causal Emotion Entailment aims to identify causal utterances that are responsible for the target utterance with a non-neutral emotion in conversations. Previous works are limited in thorough understanding of the conversational context and accurate reasoning of the emotion cause. To this end, we propose Knowledge-Bridged Causal Interaction Network (KBCIN) with commonsense knowledge (CSK) leveraged as three bridges. Specifically, we construct a conversational graph for each conversation and leverage the event-centered CSK as the semantics-level bridge (S-bridge) to capture the deep inter-utterance dependencies in the conversational context via the CSK-Enhanced Graph Attention module. Moreover, social-interaction CSK serves as emotion-level bridge (E-bridge) and action-level bridge (A-bridge) to connect candidate utterances with the target one, which provides explicit causal clues for the Emotional Interaction module and Actional Interaction module to reason the target emotion. Experimental results show that our model achieves better performance over most baseline models. Our source code is publicly available at https://github.com/circle-hit/KBCIN.

Yang Wu, Yanyan Zhao, Hao Yang et al.

Multimodal sentiment analysis has attracted increasing attention and lots of models have been proposed. However, the performance of the state-of-the-art models decreases sharply when they are deployed in the real world. We find that the main reason is that real-world applications can only access the text outputs by the automatic speech recognition (ASR) models, which may be with errors because of the limitation of model capacity. Through further analysis of the ASR outputs, we find that in some cases the sentiment words, the key sentiment elements in the textual modality, are recognized as other words, which makes the sentiment of the text change and hurts the performance of multimodal sentiment models directly. To address this problem, we propose the sentiment word aware multimodal refinement model (SWRM), which can dynamically refine the erroneous sentiment words by leveraging multimodal sentiment clues. Specifically, we first use the sentiment word position detection module to obtain the most possible position of the sentiment word in the text and then utilize the multimodal sentiment word refinement module to dynamically refine the sentiment word embeddings. The refined embeddings are taken as the textual inputs of the multimodal feature fusion module to predict the sentiment labels. We conduct extensive experiments on the real-world datasets including MOSI-Speechbrain, MOSI-IBM, and MOSI-iFlytek and the results demonstrate the effectiveness of our model, which surpasses the current state-of-the-art models on three datasets. Furthermore, our approach can be adapted for other multimodal feature fusion models easily. Data and code are available at https://github.com/albertwy/SWRM.

Haochun Wang, Chi Liu, Nuwa Xi et al.

Large Language Models (LLMs), such as the LLaMA model, have demonstrated their effectiveness in various general-domain natural language processing (NLP) tasks. Nevertheless, LLMs have not yet performed optimally in biomedical domain tasks due to the need for medical expertise in the responses. In response to this challenge, we propose HuaTuo, a LLaMA-based model that has been supervised-fine-tuned with generated QA (Question-Answer) instances. The experimental results demonstrate that HuaTuo generates responses that possess more reliable medical knowledge. Our proposed HuaTuo model is accessible at https://github.com/SCIR-HI/Huatuo-Llama-Med-Chinese.

Weixiang Zhao, Yichen Zhang, Yingshuo Wang et al. · cmu

Experience-driven self-evolution has emerged as a promising paradigm for improving the autonomy of large language model agents, yet its reliance on self-curated experience introduces underexplored safety risks. In this study, we investigate how experience accumulation and utilization in self-evolving agents affect safety performance across web-based and embodied environments. Notably, experience gathered solely from benign tasks can still compromise safety in high-risk scenarios. Further analysis attributes this degradation to the execution-oriented nature of accumulated experience, which reinforces agents' tendency to act rather than refuse. In more realistic settings where agents encounter both benign and harmful tasks, refusal-related experience mitigates safety decline but induces over-refusal, revealing a fundamental safety-utility trade-off. Overall, our findings expose inherent limitations of current self-evolving agents and call for more principled strategies to ensure safe and reliable adaptation.

HuiMing Fan, Xiao Wang, Zheng Chu et al.

Are LLM-based search agents genuinely searching, or using the web to verify what they already know? We study this question on BrowseComp with three diagnostics. Our analysis reveals Intrinsic Knowledge Dependence (IKD): even with tool access, agents often rely on intrinsic knowledge -- information encoded in the model before retrieval -- rather than on external evidence. Agents answer up to 44.5% of BrowseComp questions without tools, generate more than half of their search queries from internally produced hypotheses rather than retrieved leads, and perform worse than closed-book baselines when answer-supporting evidence is removed. These results suggest that static search benchmarks can reward memory-backed verification rather than evidence-driven discovery, conflating what agents already know with what they can find. We then introduce LiveBrowseComp, a deep-search benchmark designed to evaluate agents beyond intrinsic coverage. It contains 335 human-authored questions whose answers depend on facts published within the 90 days preceding benchmark construction, drawn from six updated sources and filtered to exclude globally salient events. On LiveBrowseComp, all evaluated agents fall below 2% closed-book accuracy, search-augmented scores drop by 25-40 points relative to BrowseComp, and prior model rankings no longer reliably predict performance. LiveBrowseComp is available at https://huggingface.co/datasets/Forival/LiveBrowseComp.

Yexing Du, Kaiyuan Liu, Youcheng Pan et al.

Multimodal large language models (MLLMs) have demonstrated significant potential for speech-to-text translation (S2TT). However, existing deployment paradigms face critical challenges: pure on-device models suffer from resource constraints, while centralized cloud systems incur severe privacy risks and bandwidth bottlenecks by transmitting raw voice data. Furthermore, most models exhibit English-centric biases, restricting many-to-many translation scaling. In this paper, we propose Edge-cloud Speech Recognition and Translation (ESRT), a privacy-preserving and bandwidth-efficient collaborative edge-cloud MLLM framework. Specifically, we design an edge-cloud split inference architecture that retains a lightweight speech encoder and adapter on the device, transmitting only highly compressed intermediate features to the cloud. This fundamentally prevents voiceprint leakage and reduces bandwidth requirements by up to 10$\times$. To overcome English-centric bottlenecks, we introduce a multi-task weighted curriculum learning strategy with data balancing to ensure robust cross-lingual consistency. Extensive experiments on the FLEURS dataset demonstrate that our models, ESRT-4B and ESRT-12B, achieve state-of-the-art many-to-many S2TT performance across 45 languages ($45 \times 44$ directions). Code and models are released to facilitate reproducible, privacy-aware MLLM S2TT research. The code and models are released at https://github.com/yxduir/esrt.

Yexing Du, Youcheng Pan, Ziyang Ma et al.

Multimodal Large Language Models (MLLMs) have achieved significant success in Speech-to-Text Translation (S2TT) tasks. While most existing research has focused on English-centric translation directions, the exploration of many-to-many translation is still limited by the scarcity of parallel data. To address this, we propose a three-stage curriculum learning strategy that leverages the machine translation capabilities of large language models and adapts them to S2TT tasks, enabling effective learning in low-resource settings. We trained MLLMs with varying parameter sizes (3B, 7B, and 32B) and evaluated the proposed strategy using the FLEURS and CoVoST-2 datasets. Experimental results show that the proposed strategy achieves state-of-the-art average performance in $15\times14$ language pairs, requiring fewer than 10 hours of speech data per language to achieve competitive results. The source code and models are released at https://github.com/yxduir/LLM-SRT.

Lei Huang, Weijiang Yu, Weitao Ma et al.

The emergence of large language models (LLMs) has marked a significant breakthrough in natural language processing (NLP), fueling a paradigm shift in information acquisition. Nevertheless, LLMs are prone to hallucination, generating plausible yet nonfactual content. This phenomenon raises significant concerns over the reliability of LLMs in real-world information retrieval (IR) systems and has attracted intensive research to detect and mitigate such hallucinations. Given the open-ended general-purpose attributes inherent to LLMs, LLM hallucinations present distinct challenges that diverge from prior task-specific models. This divergence highlights the urgency for a nuanced understanding and comprehensive overview of recent advances in LLM hallucinations. In this survey, we begin with an innovative taxonomy of hallucination in the era of LLM and then delve into the factors contributing to hallucinations. Subsequently, we present a thorough overview of hallucination detection methods and benchmarks. Our discussion then transfers to representative methodologies for mitigating LLM hallucinations. Additionally, we delve into the current limitations faced by retrieval-augmented LLMs in combating hallucinations, offering insights for developing more robust IR systems. Finally, we highlight the promising research directions on LLM hallucinations, including hallucination in large vision-language models and understanding of knowledge boundaries in LLM hallucinations.

Zheng Chu, Jingchang Chen, Qianglong Chen et al.

Grasping the concept of time is a fundamental facet of human cognition, indispensable for truly comprehending the intricacies of the world. Previous studies typically focus on specific aspects of time, lacking a comprehensive temporal reasoning benchmark. To address this, we propose TimeBench, a comprehensive hierarchical temporal reasoning benchmark that covers a broad spectrum of temporal reasoning phenomena. TimeBench provides a thorough evaluation for investigating the temporal reasoning capabilities of large language models. We conduct extensive experiments on GPT-4, LLaMA2, and other popular LLMs under various settings. Our experimental results indicate a significant performance gap between the state-of-the-art LLMs and humans, highlighting that there is still a considerable distance to cover in temporal reasoning. Besides, LLMs exhibit capability discrepancies across different reasoning categories. Furthermore, we thoroughly analyze the impact of multiple aspects on temporal reasoning and emphasize the associated challenges. We aspire for TimeBench to serve as a comprehensive benchmark, fostering research in temporal reasoning. Resources are available at: https://github.com/zchuz/TimeBench

Haochun Wang, Chi Liu, Nuwa Xi et al. · tencent-ai

Prompt-based fine-tuning for pre-trained models has proven effective for many natural language processing tasks under few-shot settings in general domain. However, tuning with prompt in biomedical domain has not been investigated thoroughly. Biomedical words are often rare in general domain, but quite ubiquitous in biomedical contexts, which dramatically deteriorates the performance of pre-trained models on downstream biomedical applications even after fine-tuning, especially in low-resource scenarios. We propose a simple yet effective approach to helping models learn rare biomedical words during tuning with prompt. Experimental results show that our method can achieve up to 6% improvement in biomedical natural language inference task without any extra parameters or training steps using few-shot vanilla prompt settings.

Li Du, Xiao Ding, Kai Xiong et al.

Understanding causality has vital importance for various Natural Language Processing (NLP) applications. Beyond the labeled instances, conceptual explanations of the causality can provide deep understanding of the causal facts to facilitate the causal reasoning process. However, such explanation information still remains absent in existing causal reasoning resources. In this paper, we fill this gap by presenting a human-annotated explainable CAusal REasoning dataset (e-CARE), which contains over 21K causal reasoning questions, together with natural language formed explanations of the causal questions. Experimental results show that generating valid explanations for causal facts still remains especially challenging for the state-of-the-art models, and the explanation information can be helpful for promoting the accuracy and stability of causal reasoning models.

Yangfan Ye, Xiaocheng Feng, Jialong Tang et al.

Existing research largely reduces cultural intelligence in LLMs to a knowledge-level problem, overlooking whether models can effectively utilize their acquired knowledge in realistic scenarios. To bridge this gap, we introduce CultureForest, a benchmark for \textit{Cultural Norm Grounded Reasoning}. Each question is grounded in a small set of atomic norms, enabling verifiable and attributable evaluation. CultureForest comprises 5,378 examples across 8 domains and 53 countries/regions, and supports a progressive evaluation from multiple-choice to open-ended generation. Extensive experiments reveal that even top-tier models degrade substantially in open-ended settings, accompanied by pronounced cross-region disparities. Through targeted analysis, we uncover several consistent patterns: (1) test-time reasoning yields limited gains and may exacerbate inequity; (2) models exhibit highly shared regional preference structures; (3) model responses are markedly conservative, especially under stricter cultural constraints; and (4) by disentangling cultural knowledge acquisition from cultural reasoning, we show that while LLMs possess substantial cultural knowledge, their performance is further bottlenecked by its effective use. These findings point to a necessary shift from knowledge-centric evaluation toward measuring knowledge-grounded reasoning.

Yuhan Chen, Nuwa Xi, Yanrui Du et al.

Molecule discovery serves as a cornerstone in numerous scientific domains, fueling the development of new materials and innovative drug designs. Recent developments of in-silico molecule discovery have highlighted the promising results of cross-modal techniques, which bridge molecular structures with their descriptive annotations. However, these cross-modal methods frequently encounter the issue of data scarcity, hampering their performance and application. In this paper, we address the low-resource challenge by utilizing artificially-real data generated by Large Language Models (LLMs). We first introduce a retrieval-based prompting strategy to construct high-quality pseudo data, then explore the optimal method to effectively leverage this pseudo data. Experiments show that using pseudo data for domain adaptation outperforms all existing methods, while also requiring a smaller model scale, reduced data size and lower training cost, highlighting its efficiency. Furthermore, our method shows a sustained improvement as the volume of pseudo data increases, revealing the great potential of pseudo data in advancing low-resource cross-modal molecule discovery. Our code and data are available at https://github.com/SCIR-HI/ArtificiallyR2R.

Yuxuan Gu, Xiaocheng Feng, Sicheng Ma et al.

Multi-aspect controllable text generation is a more challenging and practical task than single-aspect control. Existing methods achieve complex multi-aspect control by fusing multiple controllers learned from single-aspect, but suffer from attribute degeneration caused by the mutual interference of these controllers. To address this, we provide observations on attribute fusion from a distributional perspective and propose to directly search for the intersection areas of multiple attribute distributions as their combination for generation. Our method first estimates the attribute space with an autoencoder structure. Afterward, we iteratively approach the intersections by jointly minimizing distances to points representing different attributes. Finally, we map them to attribute-relevant sentences with a prefix-tuning-based decoder. Experiments on the three-aspect control task, including sentiment, topic, and detoxification aspects, reveal that our method outperforms several strong baselines on attribute relevance and text quality and achieves the SOTA. Further analysis also supplies some explanatory support for the effectiveness of our approach.

Weixiang Zhao, Yanyan Zhao, Xin Lu et al.

As a critical step to achieve human-like chatbots, empathetic response generation has attained increasing interests. Previous attempts are incomplete and not sufficient enough to elicit empathy because they only focus on the initial aspect of empathy to automatically mimic the feelings and thoughts of the user via other-awareness. However, they ignore to maintain and take the own views of the system into account, which is a crucial process to achieve the empathy called self-other awareness. To this end, we propose to generate Empathetic response with explicit Self-Other Awareness (EmpSOA). Specifically, three stages, self-other differentiation, self-other modulation and self-other generation, are devised to clearly maintain, regulate and inject the self-other aware information into the process of empathetic response generation. Both automatic and human evaluations on the benchmark dataset demonstrate the superiority of EmpSOA to generate more empathetic responses.

Yuxuan Gu, Xiaocheng Feng, Sicheng Ma et al.

Previous work on controllable text generation has explored the idea of control from the latent space, such as optimizing a representation with attribute-related classifiers or sampling a representation from relevant discrete samples. However, they are not effective enough in modeling both the latent space and the control, leaving controlled text with low quality and diversity. In this work, we propose a novel control framework using probability density estimation in the latent space. Our method utilizes an invertible transformation function, the Normalizing Flow, that maps the complex distributions in the latent space to simple Gaussian distributions in the prior space. Thus, we can perform sophisticated and flexible control in the prior space and feed the control effects back into the latent space owing to the one-one-mapping property of invertible transformations. Experiments on single-attribute controls and multi-attribute control reveal that our method outperforms several strong baselines on attribute relevance and text quality and achieves the SOTA. Further analysis of control strength adjustment demonstrates the flexibility of our control strategy.

Yuxin Wang, Minghua Ma, Zekun Wang et al.

The colossal parameters and computational overhead of Large Language Models (LLMs) challenge their real-world applications. Network pruning, which targets unstructured or structured sparsity by removing redundant parameters, has recently been explored for LLM acceleration. Existing LLM pruning works focus on unstructured pruning, which typically requires special hardware support for a practical speed-up. In contrast, structured pruning can reduce latency on general devices. However, it remains a challenge to perform structured pruning efficiently and maintain performance, especially at high sparsity ratios. To this end, we introduce an efficient structured pruning framework named CFSP, which leverages both Coarse (interblock) and Fine-grained (intrablock) activation information as an importance criterion to guide pruning. The pruning is highly efficient, as it only requires one forward pass to compute feature activations. Specifically, we first allocate the sparsity budget across blocks based on their importance and then retain important weights within each block. In addition, we introduce a recovery fine-tuning strategy that adaptively allocates training overhead based on coarse-grained importance to further improve performance. Experimental results demonstrate that CFSP outperforms existing methods on diverse models across various sparsity budgets. Our code will be available at https://github.com/wyxscir/CFSP.

Li Du, Xiao Ding, Yue Zhang et al.

Predicting the subsequent event for an existing event context is an important but challenging task, as it requires understanding the underlying relationship between events. Previous methods propose to retrieve relational features from event graph to enhance the modeling of event correlation. However, the sparsity of event graph may restrict the acquisition of relevant graph information, and hence influence the model performance. To address this issue, we consider automatically building of event graph using a BERT model. To this end, we incorporate an additional structured variable into BERT to learn to predict the event connections in the training process. Hence, in the test process, the connection relationship for unseen events can be predicted by the structured variable. Results on two event prediction tasks: script event prediction and story ending prediction, show that our approach can outperform state-of-the-art baseline methods.

Yanrui Du, Jing Yan, Yan Chen et al.

Recent research has revealed that deep neural networks often take dataset biases as a shortcut to make decisions rather than understand tasks, leading to failures in real-world applications. In this study, we focus on the spurious correlation between word features and labels that models learn from the biased data distribution of training data. In particular, we define the word highly co-occurring with a specific label as biased word, and the example containing biased word as biased example. Our analysis shows that biased examples are easier for models to learn, while at the time of prediction, biased words make a significantly higher contribution to the models' predictions, and models tend to assign predicted labels over-relying on the spurious correlation between words and labels. To mitigate models' over-reliance on the shortcut (i.e. spurious correlation), we propose a training strategy Less-Learn-Shortcut (LLS): our strategy quantifies the biased degree of the biased examples and down-weights them accordingly. Experimental results on Question Matching, Natural Language Inference and Sentiment Analysis tasks show that LLS is a task-agnostic strategy and can improve the model performance on adversarial data while maintaining good performance on in-domain data.

Zike Yuan, Ming Liu, Hui Wang et al.

Evaluating the graph comprehension and reasoning abilities of Large Language Models (LLMs) is challenging and often incomplete. Existing benchmarks focus primarily on pure graph understanding, lacking a comprehensive evaluation across all graph types and detailed capability definitions. This paper presents GraCoRe, a benchmark for systematically assessing LLMs' graph comprehension and reasoning. GraCoRe uses a three-tier hierarchical taxonomy to categorize and test models on pure graph and heterogeneous graphs, subdividing capabilities into 10 distinct areas tested through 19 tasks. Our benchmark includes 11 datasets with 5,140 graphs of varying complexity. We evaluate four closed-source and eight open-source LLMs, conducting thorough analyses from both ability and task perspectives. Key findings reveal that OpenAI o1 model has amazing comprehension and reasoning capabilities, semantic enrichment enhances reasoning performance, node ordering impacts task success, and the ability to process longer texts does not necessarily improve graph comprehension or reasoning.GraCoRe is open-sourced at https://github.com/ZIKEYUAN/GraCoRe

Zekai Ye, Qiming Li, Xiaocheng Feng et al.

While Reinforcement Learning from Verifiable Rewards (RLVR) has advanced reasoning in Large Vision-Language Models (LVLMs), prevailing frameworks suffer from a foundational methodological flaw: by distributing identical advantages across all generated tokens, these methods inherently dilute the learning signals essential for optimizing the critical, visually-grounded steps of multimodal reasoning. To bridge this gap, we formulate \textit{Token Visual Dependency}, quantifying the causal information gain of visual inputs via the Kullback-Leibler (KL) divergence between visual-conditioned and text-only predictive distributions. Revealing that this dependency is highly sparse and semantically pivotal, we introduce Perception-Grounded Policy Optimization (PGPO), which is a novel fine-grained credit assignment framework that dynamically reshapes advantages at the token level. Through a threshold-gated, mass-conserving mechanism, PGPO actively amplifies learning signals for visually-dependent tokens while suppressing gradient noise from linguistic priors. Extensive experiments based on the Qwen2.5-VL series across seven challenging multimodal reasoning benchmarks demonstrate that PGPO boosts models by 18.7% on average. Both theoretical and empirical analyses confirm that PGPO effectively reduces gradient variance, prevents training collapse, and acts as a potent regularizer for robust, perception-grounded multimodal reasoning. Code will be published on https://github.com/Yzk1114/PGPO.

Haojie Pan, Zepeng Zhai, Yuzhou Zhang et al.

Online encyclopedias, such as Wikipedia, have been well-developed and researched in the last two decades. One can find any attributes or other information of a wiki item on a wiki page edited by a community of volunteers. However, the traditional text, images and tables can hardly express some aspects of an wiki item. For example, when we talk about ``Shiba Inu'', one may care more about ``How to feed it'' or ``How to train it not to protect its food''. Currently, short-video platforms have become a hallmark in the online world. Whether you're on TikTok, Instagram, Kuaishou, or YouTube Shorts, short-video apps have changed how we consume and create content today. Except for producing short videos for entertainment, we can find more and more authors sharing insightful knowledge widely across all walks of life. These short videos, which we call knowledge videos, can easily express any aspects (e.g. hair or how-to-feed) consumers want to know about an item (e.g. Shiba Inu), and they can be systematically analyzed and organized like an online encyclopedia. In this paper, we propose Kuaipedia, a large-scale multi-modal encyclopedia consisting of items, aspects, and short videos lined to them, which was extracted from billions of videos of Kuaishou (Kwai), a well-known short-video platform in China. We first collected items from multiple sources and mined user-centered aspects from millions of users' queries to build an item-aspect tree. Then we propose a new task called ``multi-modal item-aspect linking'' as an expansion of ``entity linking'' to link short videos into item-aspect pairs and build the whole short-video encyclopedia. Intrinsic evaluations show that our encyclopedia is of large scale and highly accurate. We also conduct sufficient extrinsic experiments to show how Kuaipedia can help fundamental applications such as entity typing and entity linking.

Weixiang Zhao, Yanyan Zhao, Xin Lu et al.

This report presents a study on the emotional dialogue capability of ChatGPT, an advanced language model developed by OpenAI. The study evaluates the performance of ChatGPT on emotional dialogue understanding and generation through a series of experiments on several downstream tasks. Our findings indicate that while ChatGPT's performance on emotional dialogue understanding may still lag behind that of supervised models, it exhibits promising results in generating emotional responses. Furthermore, the study suggests potential avenues for future research directions.

Zhangyin Feng, Weitao Ma, Weijiang Yu et al.

Large language models (LLMs) exhibit superior performance on various natural language tasks, but they are susceptible to issues stemming from outdated data and domain-specific limitations. In order to address these challenges, researchers have pursued two primary strategies, knowledge editing and retrieval augmentation, to enhance LLMs by incorporating external information from different aspects. Nevertheless, there is still a notable absence of a comprehensive survey. In this paper, we propose a review to discuss the trends in integration of knowledge and large language models, including taxonomy of methods, benchmarks, and applications. In addition, we conduct an in-depth analysis of different methods and point out potential research directions in the future. We hope this survey offers the community quick access and a comprehensive overview of this research area, with the intention of inspiring future research endeavors.

Zhangyin Feng, Xiaocheng Feng, Dezhi Zhao et al.

Large language models augmented with task-relevant documents have demonstrated impressive performance on knowledge-intensive tasks. However, regarding how to obtain effective documents, the existing methods are mainly divided into two categories. One is to retrieve from an external knowledge base, and the other is to utilize large language models to generate documents. We propose an iterative retrieval-generation collaborative framework. It is not only able to leverage both parametric and non-parametric knowledge, but also helps to find the correct reasoning path through retrieval-generation interactions, which is very important for tasks that require multi-step reasoning. We conduct experiments on four question answering datasets, including single-hop QA and multi-hop QA tasks. Empirical results show that our method significantly improves the reasoning ability of large language models and outperforms previous baselines.

Kun Zhu, Xiaocheng Feng, Xiachong Feng et al.

Scientific literature review generation aims to extract and organize important information from an abundant collection of reference papers and produces corresponding reviews while lacking a clear and logical hierarchy. We observe that a high-quality catalogue-guided generation process can effectively alleviate this problem. Therefore, we present an atomic and challenging task named Hierarchical Catalogue Generation for Literature Review as the first step for review generation, which aims to produce a hierarchical catalogue of a review paper given various references. We construct a novel English Hierarchical Catalogues of Literature Reviews Dataset with 7.6k literature review catalogues and 389k reference papers. To accurately assess the model performance, we design two evaluation metrics for informativeness and similarity to ground truth from semantics and structure.Our extensive analyses verify the high quality of our dataset and the effectiveness of our evaluation metrics. We further benchmark diverse experiments on state-of-the-art summarization models like BART and large language models like ChatGPT to evaluate their capabilities. We further discuss potential directions for this task to motivate future research.

Tingting Wu, Xiao Ding, Hao Zhang et al.

Given data with label noise (i.e., incorrect data), deep neural networks would gradually memorize the label noise and impair model performance. To relieve this issue, curriculum learning is proposed to improve model performance and generalization by ordering training samples in a meaningful (e.g., easy to hard) sequence. Previous work takes incorrect samples as generic hard ones without discriminating between hard samples (i.e., hard samples in correct data) and incorrect samples. Indeed, a model should learn from hard samples to promote generalization rather than overfit to incorrect ones. In this paper, we address this problem by appending a novel loss function DiscrimLoss, on top of the existing task loss. Its main effect is to automatically and stably estimate the importance of easy samples and difficult samples (including hard and incorrect samples) at the early stages of training to improve the model performance. Then, during the following stages, DiscrimLoss is dedicated to discriminating between hard and incorrect samples to improve the model generalization. Such a training strategy can be formulated dynamically in a self-supervised manner, effectively mimicking the main principle of curriculum learning. Experiments on image classification, image regression, text sequence regression, and event relation reasoning demonstrate the versatility and effectiveness of our method, particularly in the presence of diversified noise levels.

Yanrui Du, Sendong Zhao, Yibo Gao et al.

Large language models (LLMs) equipped with chain-of-thought (CoT) achieve strong performance and offer a window into LLM behavior. However, recent evidence suggests that improvements in CoT capabilities often come with redundant reasoning processes, motivating a key question: Can LLMs acquire a fast-thinking mode analogous to human System 1 reasoning? To explore this, our study presents a self-sampling framework based on activation steering for efficient CoT learning. Our method can induce style-aligned and variable-length reasoning traces from target LLMs themselves without any teacher guidance, thereby alleviating a central bottleneck of SFT-based methods-the scarcity of high-quality supervision data. Using filtered data by gold answers, we perform SFT for efficient CoT learning with (i) a human-like dual-cognitive system, and (ii) a progressive compression curriculum. Furthermore, we explore a self-evolution regime in which SFT is driven solely by prediction-consistent data of variable-length variants, eliminating the need for gold answers. Extensive experiments on math benchmarks, together with cross-domain generalization tests in medicine, show that our method yields stable improvements for both general and R1-style LLMs. Our data and model checkpoints can be found at https://github.com/DYR1/S3-CoT.

Yanrui Du, Sendong Zhao, Muzhen Cai et al. · baidu

Extensive studies have been devoted to privatizing general-domain Large Language Models (LLMs) as Domain-Specific LLMs via feeding specific-domain data. However, these privatization efforts often ignored a critical aspect: Dual Logic Ability, which is a core reasoning ability for LLMs. The dual logic ability of LLMs ensures that they can maintain a consistent stance when confronted with both positive and negative statements about the same fact. Our study focuses on how the dual logic ability of LLMs is affected during the privatization process in the medical domain. We conduct several experiments to analyze the dual logic ability of LLMs by examining the consistency of the stance in responses to paired questions about the same fact. In our experiments, interestingly, we observed a significant decrease in the dual logic ability of existing LLMs after privatization. Besides, our results indicate that incorporating general domain dual logic data into LLMs not only enhances LLMs' dual logic ability but also further improves their accuracy. These findings underscore the importance of prioritizing LLMs' dual logic ability during the privatization process. Our study establishes a benchmark for future research aimed at exploring LLMs' dual logic ability during the privatization process and offers valuable guidance for privatization efforts in real-world applications.

Weihong Zhong, Mao Zheng, Duyu Tang et al.

Although large-scale video-language pre-training models, which usually build a global alignment between the video and the text, have achieved remarkable progress on various downstream tasks, the idea of adopting fine-grained information during the pre-training stage is not well explored. In this work, we propose STOA-VLP, a pre-training framework that jointly models object and action information across spatial and temporal dimensions. More specifically, the model regards object trajectories across frames and multiple action features from the video as fine-grained features. Besides, We design two auxiliary tasks to better incorporate both kinds of information into the pre-training process of the video-language model. The first is the dynamic object-text alignment task, which builds a better connection between object trajectories and the relevant noun tokens. The second is the spatial-temporal action set prediction, which guides the model to generate consistent action features by predicting actions found in the text. Extensive experiments on three downstream tasks (video captioning, text-video retrieval, and video question answering) demonstrate the effectiveness of our proposed STOA-VLP (e.g. 3.7 Rouge-L improvements on MSR-VTT video captioning benchmark, 2.9% accuracy improvements on MSVD video question answering benchmark, compared to previous approaches).

Xiachong Feng, Xiaocheng Feng, Bing Qin

Electroencephalography-to-Text generation (EEG-to-Text), which aims to directly generate natural text from EEG signals has drawn increasing attention in recent years due to the enormous potential for Brain-computer interfaces (BCIs). However, the remarkable discrepancy between the subject-dependent EEG representation and the semantic-dependent text representation poses a great challenge to this task. To mitigate this challenge, we devise a Curriculum Semantic-aware Contrastive Learning strategy (C-SCL), which effectively re-calibrates the subject-dependent EEG representation to the semantic-dependent EEG representation, thus reducing the discrepancy. Specifically, our C-SCL pulls semantically similar EEG representations together while pushing apart dissimilar ones. Besides, in order to introduce more meaningful contrastive pairs, we carefully employ curriculum learning to not only craft meaningful contrastive pairs but also make the learning progressively. We conduct extensive experiments on the ZuCo benchmark and our method combined with diverse models and architectures shows stable improvements across three types of metrics while achieving the new state-of-the-art. Further investigation proves not only its superiority in both the single-subject and low-resource settings but also its robust generalizability in the zero-shot setting.

Jianhua Yuan, Yanyan Zhao, Bing Qin

Stance detection models may tend to rely on dataset bias in the text part as a shortcut and thus fail to sufficiently learn the interaction between the targets and texts. Recent debiasing methods usually treated features learned by small models or big models at earlier steps as bias features and proposed to exclude the branch learning those bias features during inference. However, most of these methods fail to disentangle the ``good'' stance features and ``bad'' bias features in the text part. In this paper, we investigate how to mitigate dataset bias in stance detection. Motivated by causal effects, we leverage a novel counterfactual inference framework, which enables us to capture the dataset bias in the text part as the direct causal effect of the text on stances and reduce the dataset bias in the text part by subtracting the direct text effect from the total causal effect. We novelly model bias features as features that correlate with the stance labels but fail on intermediate stance reasoning subtasks and propose an adversarial bias learning module to model the bias more accurately. To verify whether our model could better model the interaction between texts and targets, we test our model on recently proposed test sets to evaluate the understanding of the task from various aspects. Experiments demonstrate that our proposed method (1) could better model the bias features, and (2) outperforms existing debiasing baselines on both the original dataset and most of the newly constructed test sets.

Zhenran Xu, Zifei Shan, Yuxin Li et al.

Modern Entity Linking (EL) systems entrench a popularity bias, yet there is no dataset focusing on tail and emerging entities in languages other than English. We present Hansel, a new benchmark in Chinese that fills the vacancy of non-English few-shot and zero-shot EL challenges. The test set of Hansel is human annotated and reviewed, created with a novel method for collecting zero-shot EL datasets. It covers 10K diverse documents in news, social media posts and other web articles, with Wikidata as its target Knowledge Base. We demonstrate that the existing state-of-the-art EL system performs poorly on Hansel (R@1 of 36.6% on Few-Shot). We then establish a strong baseline that scores a R@1 of 46.2% on Few-Shot and 76.6% on Zero-Shot on our dataset. We also show that our baseline achieves competitive results on TAC-KBP2015 Chinese Entity Linking task.

Xiachong Feng, Xiaocheng Feng, Xiyuan Du et al.

Meeting summarization has emerged as a promising technique for providing users with condensed summaries. However, existing work has focused on training models on centralized data, neglecting real-world scenarios where meeting data are infeasible to collect centrally, due to their sensitive nature. This gap motivates us to explore federated learning for meeting summarization. Two critical challenges impede progress. First, state-of-the-art summarizers are based on parameter-heavy pre-trained models. Exchanging such a model's parameters across clients imposes large bandwidth costs. Second, as real-world meeting data belong to various domains and are distributed across clients, they are instances of non-identically and independently distributed (non-IID). IID assumptions do not hold, which changes which forms of learning algorithms best apply. To address this, we propose Adapter-based Federated Selective Knowledge Distillation (AdaFedSelecKD) for training performant client models. Specifically, we develop an adapter-based summarization model where two adapters cooperatively facilitate learning using fewer parameters to reduce communication costs. Then, we devise a selective knowledge distillation strategy, assisting clients in robustly handling domain-focused modelling on their own data, while leveraging global parameters based on non-IID data. Extensive experiments on the QMSum benchmark demonstrate AdaFedSelecKD can achieve comparable performance with powerful centralized training methods, and shows its generalizability and robustness.

Yang Wu, Shilong Wang, Hao Yang et al.

In this paper, we evaluate different abilities of GPT-4V including visual understanding, language understanding, visual puzzle solving, and understanding of other modalities such as depth, thermal, video, and audio. To estimate GPT-4V's performance, we manually construct 656 test instances and carefully evaluate the results of GPT-4V. The highlights of our findings are as follows: (1) GPT-4V exhibits impressive performance on English visual-centric benchmarks but fails to recognize simple Chinese texts in the images; (2) GPT-4V shows inconsistent refusal behavior when answering questions related to sensitive traits such as gender, race, and age; (3) GPT-4V obtains worse results than GPT-4 (API) on language understanding tasks including general language understanding benchmarks and visual commonsense knowledge evaluation benchmarks; (4) Few-shot prompting can improve GPT-4V's performance on both visual understanding and language understanding; (5) GPT-4V struggles to find the nuances between two similar images and solve the easy math picture puzzles; (6) GPT-4V shows non-trivial performance on the tasks of similar modalities to image, such as video and thermal. Our experimental results reveal the ability and limitations of GPT-4V and we hope our paper can provide some insights into the application and research of GPT-4V.

Yanrui Du, Sendong Zhao, Yuhan Chen et al. · baidu

Pre-trained models have achieved success in Chinese Short Text Matching (STM) tasks, but they often rely on superficial clues, leading to a lack of robust predictions. To address this issue, it is crucial to analyze and mitigate the influence of superficial clues on STM models. Our study aims to investigate their over-reliance on the edit distance feature, commonly used to measure the semantic similarity of Chinese text pairs, which can be considered a superficial clue. To mitigate STM models' over-reliance on superficial clues, we propose a novel resampling training strategy called Gradually Learn Samples Containing Superficial Clue (GLS-CSC). Through comprehensive evaluations of In-Domain (I.D.), Robustness (Rob.), and Out-Of-Domain (O.O.D.) test sets, we demonstrate that GLS-CSC outperforms existing methods in terms of enhancing the robustness and generalization of Chinese STM models. Moreover, we conduct a detailed analysis of existing methods and reveal their commonality.

Hao Yang, Yanyan Zhao, Jianwei Liu et al.

Multimodal fine-grained sentiment analysis has recently attracted increasing attention due to its broad applications. However, the existing multimodal fine-grained sentiment datasets most focus on annotating the fine-grained elements in text but ignore those in images, which leads to the fine-grained elements in visual content not receiving the full attention they deserve. In this paper, we propose a new dataset, the Multimodal Aspect-Category Sentiment Analysis (MACSA) dataset, which contains more than 21K text-image pairs. The dataset provides fine-grained annotations for both textual and visual content and firstly uses the aspect category as the pivot to align the fine-grained elements between the two modalities. Based on our dataset, we propose the Multimodal ACSA task and a multimodal graph-based aligned model (MGAM), which adopts a fine-grained cross-modal fusion method. Experimental results show that our method can facilitate the baseline comparison for future research on this corpus. We will make the dataset and code publicly available.

Yichong Huang, Xiaocheng Feng, Xinwei Geng et al.

Although all-in-one-model multilingual neural machine translation (multilingual NMT) has achieved remarkable progress, the convergence inconsistency in the joint training is ignored, i.e., different language pairs reaching convergence in different epochs. This leads to the trained MNMT model over-fitting low-resource language translations while under-fitting high-resource ones. In this paper, we propose a novel training strategy named LSSD (Language-Specific Self-Distillation), which can alleviate the convergence inconsistency and help MNMT models achieve the best performance on each language pair simultaneously. Specifically, LSSD picks up language-specific best checkpoints for each language pair to teach the current model on the fly. Furthermore, we systematically explore three sample-level manipulations of knowledge transferring. Experimental results on three datasets show that LSSD obtains consistent improvements towards all language pairs and achieves the state-of-the-art.

Yang He, Xiao Ding, Bibo Cai et al.

Tool-Integrated Reasoning (TIR) extends LLM capabilities by leveraging external environments. However, existing methods lack the deliberation during sequential tool invocation required for strategic planning and self-correction. While RL mitigates this, conventional approaches for Tool-Integrated Reasoning are hindered by sparse outcome-based rewards, failing to supervise intermediate reasoning steps and tool invocations. To address this, we propose DeepTool, a novel framework that scales deliberate thinking within the interleaved process of thinking, action, and observation at each turn. In DeepTool, we first introduce a synthesis pipeline that evolves extended thinking into interleaved trajectories, integrating adversarial perturbations to ensure robustness and self-correction. Secondly, we devise Process-Supervised Reinforcement Learning based on GRPO, which utilizes an Action-Centric Process Reward to reinforce intermediate interleaved thinking and enforce precise tool invocation at every turn. Extensive experiments demonstrate that DeepTool achieves superior performance, boosting Qwen2.5-7B significantly across six benchmarks (e.g., AIME24: 3.2% -> 40.4% and HMMT25: 0.0% -> 28.6%). Furthermore, the token cost-effectiveness analysis confirms the utility of interleaved thinking, demonstrating DeepTool's optimal balance between performance and token efficiency.

Kai Xiong, Xiao Ding, Zhongyang Li et al.

Causal chain reasoning (CCR) is an essential ability for many decision-making AI systems, which requires the model to build reliable causal chains by connecting causal pairs. However, CCR suffers from two main transitive problems: threshold effect and scene drift. In other words, the causal pairs to be spliced may have a conflicting threshold boundary or scenario. To address these issues, we propose a novel Reliable Causal chain reasoning framework~(ReCo), which introduces exogenous variables to represent the threshold and scene factors of each causal pair within the causal chain, and estimates the threshold and scene contradictions across exogenous variables via structural causal recurrent neural networks~(SRNN). Experiments show that ReCo outperforms a series of strong baselines on both Chinese and English CCR datasets. Moreover, by injecting reliable causal chain knowledge distilled by ReCo, BERT can achieve better performances on four downstream causal-related tasks than BERT models enhanced by other kinds of knowledge.

Yanrui Du, Yibo Gao, Sendong Zhao et al.

R1-style LLMs have attracted growing attention for their capacity for self-reflection, yet the internal mechanisms underlying such behavior remain unclear. To bridge this gap, we anchor on the onset of reflection behavior and trace its layer-wise activation trajectory. Using the logit lens to read out token-level semantics, we uncover a structured progression: (i) Latent-control layers, where an approximate linear direction encodes the semantics of thinking budget; (ii) Semantic-pivot layers, where discourse-level cues, including turning-point and summarization cues, surface and dominate the probability mass; and (iii) Behavior-overt layers, where the likelihood of reflection-behavior tokens begins to rise until they become highly likely to be sampled. Moreover, our targeted interventions uncover a causal chain across these stages: prompt-level semantics modulate the projection of activations along latent-control directions, thereby inducing competition between turning-point and summarization cues in semantic-pivot layers, which in turn regulates the sampling likelihood of reflection-behavior tokens in behavior-overt layers. Collectively, our findings suggest a human-like meta-cognitive process-progressing from latent monitoring, to discourse-level regulation, and to finally overt self-reflection. Our analysis code can be found at https://github.com/DYR1/S3-CoT.

Xiao Han, Yuzheng Fan, Sendong Zhao et al.

Clinical decision-making agents can benefit from reusing prior decision experience. However, many memory-augmented methods store experiences as independent records without explicit relational structure, which may introduce noisy retrieval, unreliable reuse, and in some cases even hurt performance compared to direct LLM inference. We propose GSEM (Graph-based Self-Evolving Memory), a clinical memory framework that organizes clinical experiences into a dual-layer memory graph, capturing both the decision structure within each experience and the relational dependencies across experiences, and supporting applicability-aware retrieval and online feedback-driven calibration of node quality and edge weights. Across MedR-Bench and MedAgentsBench with two LLM backbones, GSEM achieves the highest average accuracy among all baselines, reaching 70.90\% and 69.24\% with DeepSeek-V3.2 and Qwen3.5-35B, respectively. Code is available at https://github.com/xhan1022/gsem.

Pengfei Deng, Jianhua Yuan, Yanyan Zhao et al.

As aspect-level sentiment labels are expensive and labor-intensive to acquire, zero-shot aspect-level sentiment classification is proposed to learn classifiers applicable to new domains without using any annotated aspect-level data. In contrast, document-level sentiment data with ratings are more easily accessible. In this work, we achieve zero-shot aspect-level sentiment classification by only using document-level reviews. Our key intuition is that the sentiment representation of a document is composed of the sentiment representations of all the aspects of that document. Based on this, we propose the AF-DSC method to explicitly model such sentiment composition in reviews. AF-DSC first learns sentiment representations for all potential aspects and then aggregates aspect-level sentiments into a document-level one to perform document-level sentiment classification. In this way, we obtain the aspect-level sentiment classifier as the by-product of the document-level sentiment classifier. Experimental results on aspect-level sentiment classification benchmarks demonstrate the effectiveness of explicit utilization of sentiment composition in document-level sentiment classification. Our model with only 30k training data outperforms previous work utilizing millions of data.

Di Wu, Yanyan Zhao, Xin Lu et al.

Defending against jailbreak attacks is crucial for the safe deployment of Large Language Models (LLMs). Recent research has attempted to improve safety by training models to reason over safety rules before responding. However, a key issue lies in determining what form of safety reasoning effectively defends against jailbreak attacks, which is difficult to explicitly design or directly obtain. To address this, we propose \textbf{STAR-S} (\textbf{S}elf-\textbf{TA}ught \textbf{R}easoning based on \textbf{S}afety rules), a framework that integrates the learning of safety rule reasoning into a self-taught loop. The core of STAR-S involves eliciting reasoning and reflection guided by safety rules, then leveraging fine-tuning to enhance safety reasoning. Repeating this process creates a synergistic cycle. Improvements in the model's reasoning and interpretation of safety rules allow it to produce better reasoning data under safety rule prompts, which is then utilized for further training. Experiments show that STAR-S effectively defends against jailbreak attacks, outperforming baselines. Code is available at: https://github.com/pikepokenew/STAR_S.git.

Yexing Du, Youcheng Pan, Zekun Wang et al.

Multimodal Large Language Models (MLLMs) have achieved notable success in enhancing translation performance by integrating multimodal information. However, existing research primarily focuses on image-guided methods, whose applicability is constrained by the scarcity of multilingual image-text pairs. The speech modality overcomes this limitation due to its natural alignment with text and the abundance of existing speech datasets, which enable scalable language coverage. In this paper, we propose a Speech-guided Machine Translation (SMT) framework that integrates speech and text as fused inputs into an MLLM to improve translation quality. To mitigate reliance on low-resource data, we introduce a Self-Evolution Mechanism. The core components of this framework include a text-to-speech model, responsible for generating synthetic speech, and an MLLM capable of classifying synthetic speech samples and iteratively optimizing itself using positive samples. Experimental results demonstrate that our framework surpasses all existing methods on the Multi30K multimodal machine translation benchmark, achieving new state-of-the-art results. Furthermore, on general machine translation datasets, particularly the FLORES-200, it achieves average state-of-the-art performance in 108 translation directions. Ablation studies on CoVoST-2 confirms that differences between synthetic and authentic speech have negligible impact on translation quality. The code and models are released at https://github.com/yxduir/LLM-SRT.

Ming Liu, Yaojia LV, Jingrun Zhang et al.

This paper presents BigCilin, the first Chinese open-domain knowledge graph with fine-grained hypernym-hyponym re-lations which are extracted automatically from multiple sources for Chinese named entities. With the fine-grained hypernym-hyponym relations, BigCilin owns flexible semantic hierarchical structure. Since the hypernym-hyponym paths are automati-cally generated and one entity may have several senses, we provide a path disambi-guation solution to map a hypernym-hyponym path of one entity to its one sense on the condition that the path and the sense express the same meaning. In order to conveniently access our BigCilin Knowle-dge graph, we provide web interface in two ways. One is that it supports querying any Chinese named entity and browsing the extracted hypernym-hyponym paths surro-unding the query entity. The other is that it gives a top-down browsing view to illust-rate the overall hierarchical structure of our BigCilin knowledge graph over some sam-pled entities.

Lei Huang, Xiaocheng Feng, Weitao Ma et al.

Despite the impressive performance on information-seeking tasks, large language models (LLMs) still struggle with hallucinations. Attributed LLMs, which augment generated text with in-line citations, have shown potential in mitigating hallucinations and improving verifiability. However, current approaches suffer from suboptimal citation quality due to their reliance on in-context learning. Furthermore, the practice of citing only coarse document identifiers makes it challenging for users to perform fine-grained verification. In this work, we introduce FRONT, a training framework designed to teach LLMs to generate Fine-Grained Grounded Citations. By grounding model outputs in fine-grained supporting quotes, these quotes guide the generation of grounded and consistent responses, not only improving citation quality but also facilitating fine-grained verification. Experiments on the ALCE benchmark demonstrate the efficacy of FRONT in generating superior grounded responses and highly supportive citations. With LLaMA-2-7B, the framework significantly outperforms all the baselines, achieving an average of 14.21% improvement in citation quality across all datasets, even surpassing ChatGPT.

Jinglong Gao, Xiao Ding, Yiming Cui et al.

To improve the performance of large language models (LLMs), researchers have explored providing LLMs with textual task-solving experience via prompts. However, they rely on manual efforts to acquire and apply such experience for each task, which is not feasible for the growing demand for LLMs and the variety of user questions. To address this issue, we design a lifelong autonomous experiential learning framework based on LLMs to explore whether LLMs can imitate human ability for learning and utilizing experience. It autonomously learns and accumulates experience through experience transfer and induction, categorizing the types of input questions to select which accumulated experience to employ for them. Experimental results on six widely used NLP datasets show that our framework performs reliably in each intermediate step and effectively improves the performance of GPT-3.5 and GPT-4. This validates the feasibility of using LLMs to mimic human experiential learning and application capabilities. Additionally, we provide a detailed analysis of the behavior of our framework at each step.

Nuwa Xi, Sendong Zhao, Haochun Wang et al.

Decoding text stimuli from cognitive signals (e.g. fMRI) enhances our understanding of the human language system, paving the way for building versatile Brain-Computer Interface. However, existing studies largely focus on decoding individual word-level fMRI volumes from a restricted vocabulary, which is far too idealized for real-world application. In this paper, we propose fMRI2text, the first openvocabulary task aiming to bridge fMRI time series and human language. Furthermore, to explore the potential of this new task, we present a baseline solution, UniCoRN: the Unified Cognitive Signal ReconstructioN for Brain Decoding. By reconstructing both individual time points and time series, UniCoRN establishes a robust encoder for cognitive signals (fMRI & EEG). Leveraging a pre-trained language model as decoder, UniCoRN proves its efficacy in decoding coherent text from fMRI series across various split settings. Our model achieves a 34.77% BLEU score on fMRI2text, and a 37.04% BLEU when generalized to EEGto-text decoding, thereby surpassing the former baseline. Experimental results indicate the feasibility of decoding consecutive fMRI volumes, and the effectiveness of decoding different cognitive signals using a unified structure.