Qingyao Ai

Qingyao Ai, Ting Bai, Zhao Cao et al. · pku, tsinghua

The research field of Information Retrieval (IR) has evolved significantly, expanding beyond traditional search to meet diverse user information needs. Recently, Large Language Models (LLMs) have demonstrated exceptional capabilities in text understanding, generation, and knowledge inference, opening up exciting avenues for IR research. LLMs not only facilitate generative retrieval but also offer improved solutions for user understanding, model evaluation, and user-system interactions. More importantly, the synergistic relationship among IR models, LLMs, and humans forms a new technical paradigm that is more powerful for information seeking. IR models provide real-time and relevant information, LLMs contribute internal knowledge, and humans play a central role of demanders and evaluators to the reliability of information services. Nevertheless, significant challenges exist, including computational costs, credibility concerns, domain-specific limitations, and ethical considerations. To thoroughly discuss the transformative impact of LLMs on IR research, the Chinese IR community conducted a strategic workshop in April 2023, yielding valuable insights. This paper provides a summary of the workshop's outcomes, including the rethinking of IR's core values, the mutual enhancement of LLMs and IR, the proposal of a novel IR technical paradigm, and open challenges.

Weihang Su, Yichen Tang, Qingyao Ai et al. · tsinghua

The emergence of Large Language Models (LLMs) has revolutionized how users access information, shifting from traditional search engines to direct question-and-answer interactions with LLMs. However, the widespread adoption of LLMs has revealed a significant challenge known as hallucination, wherein LLMs generate coherent yet factually inaccurate responses. This hallucination phenomenon has led to users' distrust in information retrieval systems based on LLMs. To tackle this challenge, this paper proposes Dynamic Retrieval Augmentation based on hallucination Detection (DRAD) as a novel method to detect and mitigate hallucinations in LLMs. DRAD improves upon traditional retrieval augmentation by dynamically adapting the retrieval process based on real-time hallucination detection. It features two main components: Real-time Hallucination Detection (RHD) for identifying potential hallucinations without external models, and Self-correction based on External Knowledge (SEK) for correcting these errors using external knowledge. Experiment results show that DRAD demonstrates superior performance in both detecting and mitigating hallucinations in LLMs. All of our code and data are open-sourced at https://github.com/oneal2000/EntityHallucination.

Kaiyuan Zhang, Ziyi Ye, Qingyao Ai et al. · tsinghua

Electroencephalography(EEG) classification is a crucial task in neuroscience, neural engineering, and several commercial applications. Traditional EEG classification models, however, have often overlooked or inadequately leveraged the brain's topological information. Recognizing this shortfall, there has been a burgeoning interest in recent years in harnessing the potential of Graph Neural Networks (GNN) to exploit the topological information by modeling features selected from each EEG channel in a graph structure. To further facilitate research in this direction, we introduce GNN4EEG, a versatile and user-friendly toolkit for GNN-based modeling of EEG signals. GNN4EEG comprises three components: (i)A large benchmark constructed with four EEG classification tasks based on EEG data collected from 123 participants. (ii)Easy-to-use implementations on various state-of-the-art GNN-based EEG classification models, e.g., DGCNN, RGNN, etc. (iii)Implementations of comprehensive experimental settings and evaluation protocols, e.g., data splitting protocols, and cross-validation protocols. GNN4EEG is publicly released at https://github.com/Miracle-2001/GNN4EEG.

Jingtao Zhan, Qingyao Ai, Yiqun Liu et al. · tsinghua

Recent advance in Dense Retrieval (DR) techniques has significantly improved the effectiveness of first-stage retrieval. Trained with large-scale supervised data, DR models can encode queries and documents into a low-dimensional dense space and conduct effective semantic matching. However, previous studies have shown that the effectiveness of DR models would drop by a large margin when the trained DR models are adopted in a target domain that is different from the domain of the labeled data. One of the possible reasons is that the DR model has never seen the target corpus and thus might be incapable of mitigating the difference between the training and target domains. In practice, unfortunately, training a DR model for each target domain to avoid domain shift is often a difficult task as it requires additional time, storage, and domain-specific data labeling, which are not always available. To address this problem, in this paper, we propose a novel DR framework named Disentangled Dense Retrieval (DDR) to support effective and flexible domain adaptation for DR models. DDR consists of a Relevance Estimation Module (REM) for modeling domain-invariant matching patterns and several Domain Adaption Modules (DAMs) for modeling domain-specific features of multiple target corpora. By making the REM and DAMs disentangled, DDR enables a flexible training paradigm in which REM is trained with supervision once and DAMs are trained with unsupervised data. Comprehensive experiments in different domains and languages show that DDR significantly improves ranking performance compared to strong DR baselines and substantially outperforms traditional retrieval methods in most scenarios.

Xuancheng Li, Tao Yang, Yujia Zhou et al.

Ranking systems in web search and recommendation allocate attention among items and providers, and therefore need to balance relevance-based effectiveness with provider fairness. Existing fair-ranking methods commonly focus on exposure fairness, where cumulative exposure is allocated in proportion to item merit. However, exposure is often only an intermediate signal: the actual utility received by a provider may depend on context-dependent conversion from exposure to income, such as clicks, purchases, or advertising value. This paper studies fair ranking under context-dependent provider utility, which we refer to as income. We formalize income fairness by requiring cumulative provider income to be proportional to relevance, and define an income-unfairness metric based on this proportionality condition. We then propose DIDRF, a Dynamic-Income-Derivative-aware Ranking Fairness algorithm for income-fair ranking. DIDRF uses the quadratic structure of income-fairness violations to derive a state-aware scoring rule that jointly considers ranking effectiveness and the marginal effect of each ranking decision on cumulative income fairness. Experiments on standard learning-to-rank datasets with log-calibrated semi-synthetic income environments based on advertising and e-commerce logs show that DIDRF consistently improves income fairness over representative fair-ranking baselines while preserving competitive ranking effectiveness.

Haitao Li, Qingyao Ai, Jia Chen et al. · tsinghua

Legal case retrieval, which aims to find relevant cases for a query case, plays a core role in the intelligent legal system. Despite the success that pre-training has achieved in ad-hoc retrieval tasks, effective pre-training strategies for legal case retrieval remain to be explored. Compared with general documents, legal case documents are typically long text sequences with intrinsic logical structures. However, most existing language models have difficulty understanding the long-distance dependencies between different structures. Moreover, in contrast to the general retrieval, the relevance in the legal domain is sensitive to key legal elements. Even subtle differences in key legal elements can significantly affect the judgement of relevance. However, existing pre-trained language models designed for general purposes have not been equipped to handle legal elements. To address these issues, in this paper, we propose SAILER, a new Structure-Aware pre-traIned language model for LEgal case Retrieval. It is highlighted in the following three aspects: (1) SAILER fully utilizes the structural information contained in legal case documents and pays more attention to key legal elements, similar to how legal experts browse legal case documents. (2) SAILER employs an asymmetric encoder-decoder architecture to integrate several different pre-training objectives. In this way, rich semantic information across tasks is encoded into dense vectors. (3) SAILER has powerful discriminative ability, even without any legal annotation data. It can distinguish legal cases with different charges accurately. Extensive experiments over publicly available legal benchmarks demonstrate that our approach can significantly outperform previous state-of-the-art methods in legal case retrieval.

Haitao Li, Yunqiu Shao, Yueyue Wu et al.

As an important component of intelligent legal systems, legal case retrieval plays a critical role in ensuring judicial justice and fairness. However, the development of legal case retrieval technologies in the Chinese legal system is restricted by three problems in existing datasets: limited data size, narrow definitions of legal relevance, and naive candidate pooling strategies used in data sampling. To alleviate these issues, we introduce LeCaRDv2, a large-scale Legal Case Retrieval Dataset (version 2). It consists of 800 queries and 55,192 candidates extracted from 4.3 million criminal case documents. To the best of our knowledge, LeCaRDv2 is one of the largest Chinese legal case retrieval datasets, providing extensive coverage of criminal charges. Additionally, we enrich the existing relevance criteria by considering three key aspects: characterization, penalty, procedure. This comprehensive criteria enriches the dataset and may provides a more holistic perspective. Furthermore, we propose a two-level candidate set pooling strategy that effectively identify potential candidates for each query case. It's important to note that all cases in the dataset have been annotated by multiple legal experts specializing in criminal law. Their expertise ensures the accuracy and reliability of the annotations. We evaluate several state-of-the-art retrieval models at LeCaRDv2, demonstrating that there is still significant room for improvement in legal case retrieval. The details of LeCaRDv2 can be found at the anonymous website https://github.com/anonymous1113243/LeCaRDv2.

Yunqiu Shao, Haitao Li, Yueyue Wu et al. · tsinghua

Legal case retrieval is a special Information Retrieval~(IR) task focusing on legal case documents. Depending on the downstream tasks of the retrieved case documents, users' information needs in legal case retrieval could be significantly different from those in Web search and traditional ad-hoc retrieval tasks. While there are several studies that retrieve legal cases based on text similarity, the underlying search intents of legal retrieval users, as shown in this paper, are more complicated than that yet mostly unexplored. To this end, we present a novel hierarchical intent taxonomy of legal case retrieval. It consists of five intent types categorized by three criteria, i.e., search for Particular Case(s), Characterization, Penalty, Procedure, and Interest. The taxonomy was constructed transparently and evaluated extensively through interviews, editorial user studies, and query log analysis. Through a laboratory user study, we reveal significant differences in user behavior and satisfaction under different search intents in legal case retrieval. Furthermore, we apply the proposed taxonomy to various downstream legal retrieval tasks, e.g., result ranking and satisfaction prediction, and demonstrate its effectiveness. Our work provides important insights into the understanding of user intents in legal case retrieval and potentially leads to better retrieval techniques in the legal domain, such as intent-aware ranking strategies and evaluation methodologies.

Haitao Li, You Chen, Qingyao Ai et al.

Large language models (LLMs) have made significant progress in natural language processing tasks and demonstrate considerable potential in the legal domain. However, legal applications demand high standards of accuracy, reliability, and fairness. Applying existing LLMs to legal systems without careful evaluation of their potential and limitations could pose significant risks in legal practice. To this end, we introduce a standardized comprehensive Chinese legal benchmark LexEval. This benchmark is notable in the following three aspects: (1) Ability Modeling: We propose a new taxonomy of legal cognitive abilities to organize different tasks. (2) Scale: To our knowledge, LexEval is currently the largest Chinese legal evaluation dataset, comprising 23 tasks and 14,150 questions. (3) Data: we utilize formatted existing datasets, exam datasets and newly annotated datasets by legal experts to comprehensively evaluate the various capabilities of LLMs. LexEval not only focuses on the ability of LLMs to apply fundamental legal knowledge but also dedicates efforts to examining the ethical issues involved in their application. We evaluated 38 open-source and commercial LLMs and obtained some interesting findings. The experiments and findings offer valuable insights into the challenges and potential solutions for developing Chinese legal systems and LLM evaluation pipelines. The LexEval dataset and leaderboard are publicly available at \url{https://github.com/CSHaitao/LexEval} and will be continuously updated.

Zhenduo Wang, Yuancheng Tu, Corby Rosset et al. · tsinghua

A long-standing challenge for search and conversational assistants is query intention detection in ambiguous queries. Asking clarifying questions in conversational search has been widely studied and considered an effective solution to resolve query ambiguity. Existing work have explored various approaches for clarifying question ranking and generation. However, due to the lack of real conversational search data, they have to use artificial datasets for training, which limits their generalizability to real-world search scenarios. As a result, the industry has shown reluctance to implement them in reality, further suspending the availability of real conversational search interaction data. The above dilemma can be formulated as a cold start problem of clarifying question generation and conversational search in general. Furthermore, even if we do have large-scale conversational logs, it is not realistic to gather training data that can comprehensively cover all possible queries and topics in open-domain search scenarios. The risk of fitting bias when training a clarifying question retrieval/generation model on incomprehensive dataset is thus another important challenge. In this work, we innovatively explore generating clarifying questions in a zero-shot setting to overcome the cold start problem and we propose a constrained clarifying question generation system which uses both question templates and query facets to guide the effective and precise question generation. The experiment results show that our method outperforms existing state-of-the-art zero-shot baselines by a large margin. Human annotations to our model outputs also indicate our method generates 25.2\% more natural questions, 18.1\% more useful questions, 6.1\% less unnatural and 4\% less useless questions.

Haitao Li, Qingyao Ai, Jingtao Zhan et al. · tsinghua

Recent studies have shown that Dense Retrieval (DR) techniques can significantly improve the performance of first-stage retrieval in IR systems. Despite its empirical effectiveness, the application of DR is still limited. In contrast to statistic retrieval models that rely on highly efficient inverted index solutions, DR models build dense embeddings that are difficult to be pre-processed with most existing search indexing systems. To avoid the expensive cost of brute-force search, the Approximate Nearest Neighbor (ANN) algorithm and corresponding indexes are widely applied to speed up the inference process of DR models. Unfortunately, while ANN can improve the efficiency of DR models, it usually comes with a significant price on retrieval performance. To solve this issue, we propose JTR, which stands for Joint optimization of TRee-based index and query encoding. Specifically, we design a new unified contrastive learning loss to train tree-based index and query encoder in an end-to-end manner. The tree-based negative sampling strategy is applied to make the tree have the maximum heap property, which supports the effectiveness of beam search well. Moreover, we treat the cluster assignment as an optimization problem to update the tree-based index that allows overlapped clustering. We evaluate JTR on numerous popular retrieval benchmarks. Experimental results show that JTR achieves better retrieval performance while retaining high system efficiency compared with widely-adopted baselines. It provides a potential solution to balance efficiency and effectiveness in neural retrieval system designs.

Changyue Wang, Weihang Su, Hu Yiran et al.

Recent advances in Large Language Models (LLMs) have significantly shaped the applications of AI in multiple fields, including the studies of legal intelligence. Trained on extensive legal texts, including statutes and legal documents, the legal LLMs can capture important legal knowledge/concepts effectively and provide important support for downstream legal applications such as legal consultancy. Yet, the dynamic nature of legal statutes and interpretations also poses new challenges to the use of LLMs in legal applications. Particularly, how to update the legal knowledge of LLMs effectively and efficiently has become an important research problem in practice. Existing benchmarks for evaluating knowledge update methods are mostly designed for the open domain and cannot address the specific challenges of the legal domain, such as the nuanced application of new legal knowledge, the complexity and lengthiness of legal regulations, and the intricate nature of legal reasoning. To address this gap, we introduce the Legal Knowledge Update BEnchmark, i.e. LeKUBE, which evaluates knowledge update methods for legal LLMs across five dimensions. Specifically, we categorize the needs of knowledge updates in the legal domain with the help of legal professionals, and then hire annotators from law schools to create synthetic updates to the Chinese Criminal and Civil Code as well as sets of questions of which the answers would change after the updates. Through a comprehensive evaluation of state-of-the-art knowledge update methods, we reveal a notable gap between existing knowledge update methods and the unique needs of the legal domain, emphasizing the need for further research and development of knowledge update mechanisms tailored for legal LLMs.

Hexi Wang, Yujia Zhou, Bangde Du et al.

Large language models (LLMs) have recently been adopted as synthetic agents for public opinion simulation, offering a promising alternative to costly and slow human surveys. Despite their scalability, current LLM-based simulation methods fail to capture social diversity, producing flattened inter-group differences and overly homogeneous responses within demographic groups. We identify this limitation as a Diversity Collapse phenomenon in LLM hidden representations, where distinct social identities become increasingly indistinguishable across layers. Motivated by this observation, we propose Parametric Social Identity Injection (PSII), a general framework that injects explicit, parametric representations of demographic attributes and value orientations directly into intermediate hidden states of LLMs. Unlike prompt-based persona conditioning, PSII enables fine-grained and controllable identity modulation at the representation level. Extensive experiments on the World Values Survey using multiple open-source LLMs show that PSII significantly improves distributional fidelity and diversity, reducing KL divergence to real-world survey data while enhancing overall diversity. This work provides new insights into representation-level control of LLM agents and advances scalable, diversity-aware public opinion simulation. Code and data are available at https://github.com/halsayxi/PSII.

Changyue Wang, Weihang Su, Qingyao Ai et al.

Scaling training data and model parameters has long driven progress in large language models (LLMs), but this paradigm is increasingly constrained by the scarcity of high-quality data and diminishing returns from rising computational costs. As a result, recent work is increasing the focus on continual learning from real-world deployment, where user interaction logs provide a rich source of authentic human feedback and procedural knowledge. However, learning from user logs is challenging due to their unstructured and noisy nature. Vanilla LLM systems often struggle to distinguish useful feedback signals from noisy user behavior, and the disparity between user log collection and model optimization (e.g., the off-policy optimization problem) further strengthens the problem. To this end, we propose UNO (User log-driveN Optimization), a unified framework for improving LLM systems (LLMsys) with user logs. UNO first distills logs into semi-structured rules and preference pairs, then employs query-and-feedback-driven clustering to manage data heterogeneity, and finally quantifies the cognitive gap between the model's prior knowledge and the log data. This assessment guides the LLMsys to adaptively filter out noisy feedback and construct different modules for primary and reflective experiences extracted from user logs, thereby improving future responses. Extensive experiments show that UNO achieves state-of-the-art effectiveness and efficiency, significantly outperforming Retrieval Augmented Generation (RAG) and memory-based baselines. We have open-sourced our code at https://github.com/bebr2/UNO .

Weihang Su, Jianming Long, Changyue Wang et al.

Large Language Models (LLMs) frequently exhibit hallucinations, generating content that appears fluent and coherent but is factually incorrect. Such errors undermine trust and hinder their adoption in real-world applications. To address this challenge, two distinct research paradigms have emerged: model-centric Hallucination Detection (HD) and text-centric Fact Verification (FV). Despite sharing the same goal, these paradigms have evolved in isolation, using distinct assumptions, datasets, and evaluation protocols. This separation has created a research schism that hinders their collective progress. In this work, we take a decisive step toward bridging this divide. We introduce UniFact, a unified evaluation framework that enables direct, instance-level comparison between FV and HD by dynamically generating model outputs and corresponding factuality labels. Through large-scale experiments across multiple LLM families and detection methods, we reveal three key findings: (1) No paradigm is universally superior; (2) HD and FV capture complementary facets of factual errors; and (3) hybrid approaches that integrate both methods consistently achieve state-of-the-art performance. Beyond benchmarking, we provide the first in-depth analysis of why FV and HD diverged, as well as empirical evidence supporting the need for their unification. The comprehensive experimental results call for a new, integrated research agenda toward unifying Hallucination Detection and Fact Verification in LLMs. We have open-sourced all the code, data, and baseline implementation at: https://github.com/oneal2000/UniFact/

Xinkai Zhang, Jingtao Zhan, Yiqun Liu et al. · tsinghua

Trial-and-error is a fundamental strategy for humans to solve complex problems and a necessary capability for Artificial Intelligence (AI) systems operating in real-world environments. Although several trial-and-error AI techniques have recently been proposed, most of them rely on simple heuristics designed by researchers and achieve limited performance gains. The core issue is the absence of appropriate data: current models cannot learn from detailed records of how humans actually conduct trial-and-error in practice. To address this gap, we introduce a data annotation platform and a corresponding dataset, termed Trial-and-Error Collection (TEC). The platform records users' complete trajectories across multiple trials and collects their reflections after receiving error feedback. Using this platform, we record the problem-solving processes of 46 participants on 58 tasks, resulting in 5,370 trial trajectories along with error reflections across 41,229 webpages. With this dataset, we observe that humans achieve substantially higher accuracy compared to LLMs, which demonstrates that humans are more effective in trial-and-error than LLMs. We believe that the TEC platform and dataset provide a valuable foundation for understanding human trial-and-error behavior and for developing more capable AI systems. Platform and dataset are publicly available.

Haitao Li, Qian Dong, Junjie Chen et al.

The rapid advancement of Large Language Models (LLMs) has driven their expanding application across various fields. One of the most promising applications is their role as evaluators based on natural language responses, referred to as ''LLMs-as-judges''. This framework has attracted growing attention from both academia and industry due to their excellent effectiveness, ability to generalize across tasks, and interpretability in the form of natural language. This paper presents a comprehensive survey of the LLMs-as-judges paradigm from five key perspectives: Functionality, Methodology, Applications, Meta-evaluation, and Limitations. We begin by providing a systematic definition of LLMs-as-Judges and introduce their functionality (Why use LLM judges?). Then we address methodology to construct an evaluation system with LLMs (How to use LLM judges?). Additionally, we investigate the potential domains for their application (Where to use LLM judges?) and discuss methods for evaluating them in various contexts (How to evaluate LLM judges?). Finally, we provide a detailed analysis of the limitations of LLM judges and discuss potential future directions. Through a structured and comprehensive analysis, we aim aims to provide insights on the development and application of LLMs-as-judges in both research and practice. We will continue to maintain the relevant resource list at https://github.com/CSHaitao/Awesome-LLMs-as-Judges.

Weihang Su, Yichen Tang, Qingyao Ai et al. · tsinghua

Dynamic retrieval augmented generation (RAG) paradigm actively decides when and what to retrieve during the text generation process of Large Language Models (LLMs). There are two key elements of this paradigm: identifying the optimal moment to activate the retrieval module (deciding when to retrieve) and crafting the appropriate query once retrieval is triggered (determining what to retrieve). However, current dynamic RAG methods fall short in both aspects. Firstly, the strategies for deciding when to retrieve often rely on static rules. Moreover, the strategies for deciding what to retrieve typically limit themselves to the LLM's most recent sentence or the last few tokens, while the LLM's real-time information needs may span across the entire context. To overcome these limitations, we introduce a new framework, DRAGIN, i.e., Dynamic Retrieval Augmented Generation based on the real-time Information Needs of LLMs. Our framework is specifically designed to make decisions on when and what to retrieve based on the LLM's real-time information needs during the text generation process. We evaluate DRAGIN along with existing methods comprehensively over 4 knowledge-intensive generation datasets. Experimental results show that DRAGIN achieves superior performance on all tasks, demonstrating the effectiveness of our method. We have open-sourced all the code, data, and models in GitHub: https://github.com/oneal2000/DRAGIN/tree/main

Xuancheng Li, Haitao Li, Yujia Zhou et al.

Long-context inputs in large language models (LLMs) often suffer from the "lost in the middle" problem, where critical information becomes diluted or ignored due to excessive length. Context compression methods aim to address this by reducing input size, but existing approaches struggle with balancing information preservation and compression efficiency. We propose Adaptive Task-Aware Compressor (ATACompressor), which dynamically adjusts compression based on the specific requirements of the task. ATACompressor employs a selective encoder that compresses only the task-relevant portions of long contexts, ensuring that essential information is preserved while reducing unnecessary content. Its adaptive allocation controller perceives the length of relevant content and adjusts the compression rate accordingly, optimizing resource utilization. We evaluate ATACompressor on three QA datasets: HotpotQA, MSMARCO, and SQUAD-showing that it outperforms existing methods in terms of both compression efficiency and task performance. Our approach provides a scalable solution for long-context processing in LLMs. Furthermore, we perform a range of ablation studies and analysis experiments to gain deeper insights into the key components of ATACompressor.

Haitao Li, Junjie Chen, Jingli Yang et al.

With the increasing intelligence and autonomy of LLM agents, their potential applications in the legal domain are becoming increasingly apparent. However, existing general-domain benchmarks cannot fully capture the complexity and subtle nuances of real-world judicial cognition and decision-making. Therefore, we propose LegalAgentBench, a comprehensive benchmark specifically designed to evaluate LLM Agents in the Chinese legal domain. LegalAgentBench includes 17 corpora from real-world legal scenarios and provides 37 tools for interacting with external knowledge. We designed a scalable task construction framework and carefully annotated 300 tasks. These tasks span various types, including multi-hop reasoning and writing, and range across different difficulty levels, effectively reflecting the complexity of real-world legal scenarios. Moreover, beyond evaluating final success, LegalAgentBench incorporates keyword analysis during intermediate processes to calculate progress rates, enabling more fine-grained evaluation. We evaluated eight popular LLMs, highlighting the strengths, limitations, and potential areas for improvement of existing models and methods. LegalAgentBench sets a new benchmark for the practical application of LLMs in the legal domain, with its code and data available at \url{https://github.com/CSHaitao/LegalAgentBench}.

Xuancheng Li, Haitao Li, Yujia Zhou et al.

Large language models (LLMs) are largely static and often redo reasoning or repeat mistakes. Prior experience reuse typically relies on external retrieval, which is similarity-based, can introduce noise, and adds latency. We introduce SEAM (Structured Experience Adapter Module), a lightweight, executor-specific plug-in that stores experience in its parameters and generates a structured, instance-tailored experience entry in a single forward pass to guide a frozen LLM executor. SEAM is trained for utility via executor rollouts and GRPO while keeping the executor frozen, and it can be further improved after deployment with supervised fine-tuning on logged successful trajectories. Experiments on mathematical reasoning benchmarks show consistent accuracy gains across executors with low overhead. Extensive ablations and analyses further elucidate the mechanisms underlying SEAM's effectiveness and robustness.

Xuancheng Li, Haitao Li, Yujia Zhou et al.

Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning in multiple domains, yet outcome-only scalar rewards are often sparse and uninformative, especially on failed samples, where they merely indicate failure and provide no insight into why the reasoning fails. In this paper, we investigate how to leverage richer verbal feedback to guide RLVR training on failed samples, and how to convert such feedback into a trainable learning signal. Specifically, we propose a multi-turn feedback-guided reinforcement learning framework. It builds on three mechanisms: (1) dynamic multi-turn regeneration guided by feedback, triggered only on failed samples, (2) two complementary learning signals for within-turn and cross-turn optimization, and (3) structured feedback injection into the model's reasoning process. Trained on sampled OpenR1-Math, the approach outperforms supervised fine-tuning and RLVR baselines in-domain and generalizes well out-of-domain.

Yiran Hu, Huanghai Liu, Chong Wang et al.

Large language models (LLMs) are being increasingly integrated into legal applications, including judicial decision support, legal practice assistance, and public-facing legal services. While LLMs show strong potential in handling legal knowledge and tasks, their deployment in real-world legal settings raises critical concerns beyond surface-level accuracy, involving the soundness of legal reasoning processes and trustworthy issues such as fairness and reliability. Systematic evaluation of LLM performance in legal tasks has therefore become essential for their responsible adoption. This survey identifies key challenges in evaluating LLMs for legal tasks grounded in real-world legal practice. We analyze the major difficulties involved in assessing LLM performance in the legal domain, including outcome correctness, reasoning reliability, and trustworthiness. Building on these challenges, we review and categorize existing evaluation methods and benchmarks according to their task design, datasets, and evaluation metrics. We further discuss the extent to which current approaches address these challenges, highlight their limitations, and outline future research directions toward more realistic, reliable, and legally grounded evaluation frameworks for LLMs in legal domains.

Yichen Tang, Weihang Su, Yiqun Liu et al.

Integrating external tools enables Large Language Models (LLMs) to interact with real-world environments and solve complex tasks. Given the growing scale of available tools, effective tool retrieval is essential to mitigate constraints of LLMs' context windows and ensure computational efficiency. Existing approaches typically treat tool retrieval as a traditional ad-hoc retrieval task, matching user queries against the entire raw tool documentation. In this paper, we identify three fundamental challenges that limit the effectiveness of this paradigm: (i) the incompleteness and structural inconsistency of tool documentation; (ii) the significant semantic and granular mismatch between user queries and technical tool documents; and, most importantly, (iii) the multi-aspect nature of tool utility, that involves distinct dimensions, such as functionality, input constraints, and output formats, varying in format and importance. To address these challenges, we introduce Multi-Field Tool Retrieval, a framework designed to align user intent with tool representations through fine-grained, multi-field modeling. Experimental results show that our framework achieves SOTA performance on five datasets and a mixed benchmark, exhibiting superior generalizability and robustness.

Yiteng Tu, Shuo Miao, Weihang Su et al.

Analytical information needs, such as trend analysis and causal impact assessment, are prevalent across various domains including law, finance, science, and much more. However, existing information retrieval paradigms, whether based on relevance-oriented document ranking or retrieval-augmented generation (RAG) with large language models (LLMs), often struggle to meet the end-to-end requirements of such tasks at the corpus scale. They either emphasize information finding rather than end-to-end problem solving, or simply treat everything as naive question answering, offering limited control over reasoning, evidence usage, and verifiability. As a result, they struggle to support analytical queries that have diverse utility concepts and high accountability requirements. In this paper, we propose analytical search as a distinct and emerging search paradigm designed to fulfill these analytical information needs. Analytical search reframes search as an evidence-governed, process-oriented analytical workflow that explicitly models analytical intent, retrieves evidence for fusion, and produces verifiable conclusions through structured, multi-step inference. We position analytical search in contrast to existing paradigms, and present a unified system framework that integrates query understanding, recall-oriented retrieval, reasoning-aware fusion, and adaptive verification. We also discuss potential research directions for the construction of analytical search engines. In this way, we highlight the conceptual significance and practical importance of analytical search and call on efforts toward the next generation of search engines that support analytical information needs.

Weihang Su, Yichen Tang, Qingyao Ai et al.

Retrieval-augmented generation (RAG) techniques have emerged as a promising solution to enhance the reliability of large language models (LLMs) by addressing issues like hallucinations, outdated knowledge, and domain adaptation. In particular, existing RAG methods append relevant documents retrieved from external corpus or databases to the input of LLMs to guide their generation process, which we refer to as the in-context knowledge injection method. While this approach is simple and often effective, it has inherent limitations. Firstly, increasing the context length and number of relevant documents can lead to higher computational overhead and degraded performance, especially in complex reasoning tasks. More importantly, in-context knowledge injection operates primarily at the input level, but LLMs store their internal knowledge in their parameters. This gap fundamentally limits the capacity of in-context methods. To this end, we introduce Parametric retrieval-augmented generation (Parametric RAG), a new RAG paradigm that integrates external knowledge directly into the parameters of feed-forward networks (FFN) of an LLM through document parameterization. This approach not only saves online computational costs by eliminating the need to inject multiple documents into the LLMs' input context, but also deepens the integration of external knowledge into the parametric knowledge space of the LLM. Experimental results demonstrate that Parametric RAG substantially enhances both the effectiveness and efficiency of knowledge augmentation in LLMs. Also, it can be combined with in-context RAG methods to achieve even better performance. We have open-sourced all the code, data, and models in the following anonymized GitHub link: https://github.com/oneal2000/PRAG

Changyue Wang, Weihang Su, Qingyao Ai et al.

Knowledge Editing is a technique that updates large language models (LLMs) with new information to maintain their world knowledge. This approach avoids the need to rebuild the model from scratch, thereby addressing the high costs associated with frequent retraining. Among these, the in-context editing paradigm stands out for its effectiveness in integrating new knowledge while preserving the model's original capabilities. Despite its potential, existing in-context knowledge editing methods are often task-specific, focusing primarily on multi-hop QA tasks using structured knowledge triples. Moreover, their reliance on few-shot prompting for task decomposition makes them unstable and less effective in generalizing across diverse tasks. In response to these limitations, we propose EditCoT, a novel knowledge editing framework that flexibly and efficiently updates LLMs across various tasks without retraining. EditCoT works by generating a chain-of-thought (CoT) for a given input and then iteratively refining this CoT process using a CoT editor based on updated knowledge. We evaluate EditCoT across a diverse range of benchmarks, covering multiple languages and tasks. The results demonstrate that our approach achieves state-of-the-art performance while offering superior generalization, effectiveness, and stability compared to existing methods, marking a significant advancement in the field of knowledge updating. The code and data of EditCoT are available at: https://github.com/bebr2/EditCoT .

Weihang Su, Baoqing Yue, Qingyao Ai et al.

This paper introduces JuDGE (Judgment Document Generation Evaluation), a novel benchmark for evaluating the performance of judgment document generation in the Chinese legal system. We define the task as generating a complete legal judgment document from the given factual description of the case. To facilitate this benchmark, we construct a comprehensive dataset consisting of factual descriptions from real legal cases, paired with their corresponding full judgment documents, which serve as the ground truth for evaluating the quality of generated documents. This dataset is further augmented by two external legal corpora that provide additional legal knowledge for the task: one comprising statutes and regulations, and the other consisting of a large collection of past judgment documents. In collaboration with legal professionals, we establish a comprehensive automated evaluation framework to assess the quality of generated judgment documents across various dimensions. We evaluate various baseline approaches, including few-shot in-context learning, fine-tuning, and a multi-source retrieval-augmented generation (RAG) approach, using both general and legal-domain LLMs. The experimental results demonstrate that, while RAG approaches can effectively improve performance in this task, there is still substantial room for further improvement. All the codes and datasets are available at: https://github.com/oneal2000/JuDGE.

Haitao Li, Yifan Chen, Yiran Hu et al.

Retrieval-augmented generation (RAG) has proven highly effective in improving large language models (LLMs) across various domains. However, there is no benchmark specifically designed to assess the effectiveness of RAG in the legal domain, which restricts progress in this area. To fill this gap, we propose LexRAG, the first benchmark to evaluate RAG systems for multi-turn legal consultations. LexRAG consists of 1,013 multi-turn dialogue samples and 17,228 candidate legal articles. Each sample is annotated by legal experts and consists of five rounds of progressive questioning. LexRAG includes two key tasks: (1) Conversational knowledge retrieval, requiring accurate retrieval of relevant legal articles based on multi-turn context. (2) Response generation, focusing on producing legally sound answers. To ensure reliable reproducibility, we develop LexiT, a legal RAG toolkit that provides a comprehensive implementation of RAG system components tailored for the legal domain. Additionally, we introduce an LLM-as-a-judge evaluation pipeline to enable detailed and effective assessment. Through experimental analysis of various LLMs and retrieval methods, we reveal the key limitations of existing RAG systems in handling legal consultation conversations. LexRAG establishes a new benchmark for the practical application of RAG systems in the legal domain, with its code and data available at https://github.com/CSHaitao/LexRAG.

Changyue Wang, Weihang Su, Qingyao Ai et al.

Large Reasoning Models (LRMs) extend large language models with explicit, multi-step reasoning traces to enhance transparency and performance on complex tasks. However, these reasoning traces can be redundant or logically inconsistent, becoming a new and hard-to-detect source of hallucination. Existing hallucination detection methods focus primarily on answer-level uncertainty and often fail to detect hallucinations or logical inconsistencies arising from the model's reasoning trace. This oversight is particularly problematic for LRMs, where the explicit thinking trace is not only an important support to the model's decision-making process but also a key source of potential hallucination. To this end, we propose RACE (Reasoning and Answer Consistency Evaluation), a novel framework specifically tailored for hallucination detection in LRMs. RACE operates by extracting essential reasoning steps and computing four diagnostic signals: inter-sample consistency of reasoning traces, entropy-based answer uncertainty, semantic alignment between reasoning and answers, and internal coherence of reasoning. The joint utilization of these signals makes RACE a more robust detector of hallucinations in LRMs. Experiments across datasets and different LLMs demonstrate that RACE outperforms existing hallucination detection baselines, offering a robust and generalizable solution for evaluating LRMs. The source code is available at https://github.com/bebr2/RACE

Changyue Wang, Weihang Su, Qingyao Ai et al.

Knowledge editing aims to efficiently update Large Language Models (LLMs) by modifying specific knowledge without retraining the entire model. Among knowledge editing approaches, in-context editing (ICE) offers a lightweight solution by injecting new knowledge directly into the input context, leaving model parameters unchanged. However, existing ICE approaches do not explicitly separate the newly injected knowledge from the model's original reasoning process. This entanglement often results in conflicts between external updates and internal parametric knowledge, undermining the consistency and accuracy of the reasoning path.In this work, we conduct preliminary experiments to examine how parametric knowledge influences reasoning path planning. We find that the model's reasoning is tightly coupled with its internal knowledge, and that naively injecting new information without adapting the reasoning path often leads to performance degradation, particularly in multi-hop tasks. To this end, we propose DecKER, a novel ICE framework that decouples reasoning from knowledge editing by generating a masked reasoning path and then resolving knowledge edits via hybrid retrieval and model-based validation. Experiments on multi-hop QA benchmarks show that DecKER significantly outperforms existing ICE methods by mitigating knowledge conflicts and preserving reasoning consistency. Our code is available at: https://github.com/bebr2/DecKER .

Haitao Li, Jiaying Ye, Yiran Hu et al.

Legal case documents play a critical role in judicial proceedings. As the number of cases continues to rise, the reliance on manual drafting of legal case documents is facing increasing pressure and challenges. The development of large language models (LLMs) offers a promising solution for automating document generation. However, existing benchmarks fail to fully capture the complexities involved in drafting legal case documents in real-world scenarios. To address this gap, we introduce CaseGen, the benchmark for multi-stage legal case documents generation in the Chinese legal domain. CaseGen is based on 500 real case samples annotated by legal experts and covers seven essential case sections. It supports four key tasks: drafting defense statements, writing trial facts, composing legal reasoning, and generating judgment results. To the best of our knowledge, CaseGen is the first benchmark designed to evaluate LLMs in the context of legal case document generation. To ensure an accurate and comprehensive evaluation, we design the LLM-as-a-judge evaluation framework and validate its effectiveness through human annotations. We evaluate several widely used general-domain LLMs and legal-specific LLMs, highlighting their limitations in case document generation and pinpointing areas for potential improvement. This work marks a step toward a more effective framework for automating legal case documents drafting, paving the way for the reliable application of AI in the legal field. The dataset and code are publicly available at https://github.com/CSHaitao/CaseGen.

Shuqi Zhu, Yi Zhong, Ziyi Ye et al.

While AI-generated hallucinations pose considerable risks, the underlying cognitive mechanisms by which humans can successfully recognize or be misled by these hallucinations remain unclear. To address this problem, this paper explores humans' neural dynamics to characterize how the brain processes hallucinated content. We record EEG signals from 27 participants while they are performing a verification task to judge the correctness of image descriptions generated by a multi-modal large language model (MLLM). Based on an averaged event-related potential (ERP) study, we reveal that multiple cognitive processes, e.g., semantic integration, inferential processing, memory retrieval, and cognitive load, exhibit distinct patterns when humans process hallucinated versus non-hallucinated content. Notably, neural responses to hallucinations that were misjudged versus correctly judged by human participants showed significant differences. This indicates that misjudged AI-generated hallucinations failed to trigger the standard neurocognitive fact verification pathway.

Zongyue Xue, Siyuan Zheng, Shaochun Wang et al.

The integration of Large Language Models (LLMs) into legal practice raises pressing concerns about judicial fairness, particularly due to the nature of their "black-box" processes. This study introduces JustEva, a comprehensive, open-source evaluation toolkit designed to measure LLM fairness in legal tasks. JustEva features several advantages: (1) a structured label system covering 65 extra-legal factors; (2) three core fairness metrics - inconsistency, bias, and imbalanced inaccuracy; (3) robust statistical inference methods; and (4) informative visualizations. The toolkit supports two types of experiments, enabling a complete evaluation workflow: (1) generating structured outputs from LLMs using a provided dataset, and (2) conducting statistical analysis and inference on LLMs' outputs through regression and other statistical methods. Empirical application of JustEva reveals significant fairness deficiencies in current LLMs, highlighting the lack of fair and trustworthy LLM legal tools. JustEva offers a convenient tool and methodological foundation for evaluating and improving algorithmic fairness in the legal domain.

Weihang Su, Anzhe Xie, Qingyao Ai et al.

The rapid growth of academic literature makes the manual creation of scientific surveys increasingly infeasible. While large language models show promise for automating this process, progress in this area is hindered by the absence of standardized benchmarks and evaluation protocols. To bridge this critical gap, we introduce SurGE (Survey Generation Evaluation), a new benchmark for scientific survey generation in computer science. SurGE consists of (1) a collection of test instances, each including a topic description, an expert-written survey, and its full set of cited references, and (2) a large-scale academic corpus of over one million papers. In addition, we propose an automated evaluation framework that measures the quality of generated surveys across four dimensions: comprehensiveness, citation accuracy, structural organization, and content quality. Our evaluation of diverse LLM-based methods demonstrates a significant performance gap, revealing that even advanced agentic frameworks struggle with the complexities of survey generation and highlighting the need for future research in this area. We have open-sourced all the code, data, and models at: https://github.com/oneal2000/SurGE

Weihang Su, Yiran Hu, Anzhe Xie et al.

Statute retrieval aims to find relevant statutory articles for specific queries. This process is the basis of a wide range of legal applications such as legal advice, automated judicial decisions, legal document drafting, etc. Existing statute retrieval benchmarks focus on formal and professional queries from sources like bar exams and legal case documents, thereby neglecting non-professional queries from the general public, which often lack precise legal terminology and references. To address this gap, we introduce the STAtute Retrieval Dataset (STARD), a Chinese dataset comprising 1,543 query cases collected from real-world legal consultations and 55,348 candidate statutory articles. Unlike existing statute retrieval datasets, which primarily focus on professional legal queries, STARD captures the complexity and diversity of real queries from the general public. Through a comprehensive evaluation of various retrieval baselines, we reveal that existing retrieval approaches all fall short of these real queries issued by non-professional users. The best method only achieves a Recall@100 of 0.907, suggesting the necessity for further exploration and additional research in this area. All the codes and datasets are available at: https://github.com/oneal2000/STARD/tree/main

Haitao Li, Changyue Wang, Weihang Su et al.

This paper describes the approach of the THUIR team at the COLIEE 2023 Legal Case Entailment task. This task requires the participant to identify a specific paragraph from a given supporting case that entails the decision for the query case. We try traditional lexical matching methods and pre-trained language models with different sizes. Furthermore, learning-to-rank methods are employed to further improve performance. However, learning-to-rank is not very robust on this task. which suggests that answer passages cannot simply be determined with information retrieval techniques. Experimental results show that more parameters and legal knowledge contribute to the legal case entailment task. Finally, we get the third place in COLIEE 2023. The implementation of our method can be found at https://github.com/CSHaitao/THUIR-COLIEE2023.

Haitao Li, Weihang Su, Changyue Wang et al.

Legal case retrieval techniques play an essential role in modern intelligent legal systems. As an annually well-known international competition, COLIEE is aiming to achieve the state-of-the-art retrieval model for legal texts. This paper summarizes the approach of the championship team THUIR in COLIEE 2023. To be specific, we design structure-aware pre-trained language models to enhance the understanding of legal cases. Furthermore, we propose heuristic pre-processing and post-processing approaches to reduce the influence of irrelevant messages. In the end, learning-to-rank methods are employed to merge features with different dimensions. Experimental results demonstrate the superiority of our proposal. Official results show that our run has the best performance among all submissions. The implementation of our method can be found at https://github.com/CSHaitao/THUIR-COLIEE2023.

Weihang Su, Changyue Wang, Qingyao Ai et al. · tsinghua

Hallucinations in large language models (LLMs) refer to the phenomenon of LLMs producing responses that are coherent yet factually inaccurate. This issue undermines the effectiveness of LLMs in practical applications, necessitating research into detecting and mitigating hallucinations of LLMs. Previous studies have mainly concentrated on post-processing techniques for hallucination detection, which tend to be computationally intensive and limited in effectiveness due to their separation from the LLM's inference process. To overcome these limitations, we introduce MIND, an unsupervised training framework that leverages the internal states of LLMs for real-time hallucination detection without requiring manual annotations. Additionally, we present HELM, a new benchmark for evaluating hallucination detection across multiple LLMs, featuring diverse LLM outputs and the internal states of LLMs during their inference process. Our experiments demonstrate that MIND outperforms existing state-of-the-art methods in hallucination detection.

Weihang Su, Jianming Long, Qingyao Ai et al.

As large language models (LLMs) evolve into agentic problem solvers, they increasingly rely on external, reusable skills to handle tasks beyond their native parametric capabilities. In existing agent systems, the dominant strategy for incorporating skills is to explicitly enumerate available skills within the context window. However, this strategy fails to scale: as skill corpora expand, context budgets are consumed rapidly, and the agent becomes markedly less accurate in identifying the right skill. To this end, this paper formulates Skill Retrieval Augmentation (SRA), a new paradigm in which agents dynamically retrieve, incorporate, and apply relevant skills from large external skill corpora on demand. To make this problem measurable, we construct a large-scale skill corpus and introduce SRA-Bench, the first benchmark for decomposed evaluation of the full SRA pipeline, covering skill retrieval, skill incorporation, and end-task execution. SRA-Bench contains 5,400 capability-intensive test instances and 636 manually constructed gold skills, which are mixed with web-collected distractor skills to form a large-scale corpus of 26,262 skills. Extensive experiments show that retrieval-based skill augmentation can substantially improve agent performance, validating the promise of the paradigm. At the same time, we uncover a fundamental gap in skill incorporation: current LLM agents tend to load skills at similar rates, regardless of whether a gold skill is retrieved or whether the task actually requires external capabilities. This shows that the bottleneck in skill augmentation lies not only in retrieval but also in the base model's ability to determine which skill to load and when external loading is actually needed. These findings position SRA as a distinct research problem and establish a foundation for the scalable augmentation of capabilities in future agent systems.

Yan Fang, Jingtao Zhan, Qingyao Ai et al. · tsinghua

Scaling up neural models has yielded significant advancements in a wide array of tasks, particularly in language generation. Previous studies have found that the performance of neural models frequently adheres to predictable scaling laws, correlated with factors such as training set size and model size. This insight is invaluable, especially as large-scale experiments grow increasingly resource-intensive. Yet, such scaling law has not been fully explored in dense retrieval due to the discrete nature of retrieval metrics and complex relationships between training data and model sizes in retrieval tasks. In this study, we investigate whether the performance of dense retrieval models follows the scaling law as other neural models. We propose to use contrastive log-likelihood as the evaluation metric and conduct extensive experiments with dense retrieval models implemented with different numbers of parameters and trained with different amounts of annotated data. Results indicate that, under our settings, the performance of dense retrieval models follows a precise power-law scaling related to the model size and the number of annotations. Additionally, we examine scaling with prevalent data augmentation methods to assess the impact of annotation quality, and apply the scaling law to find the best resource allocation strategy under a budget constraint. We believe that these insights will significantly contribute to understanding the scaling effect of dense retrieval models and offer meaningful guidance for future research endeavors.

Zhumin Chu, Qingyao Ai, Yiteng Tu et al.

The impressive performance of large language models (LLMs) has attracted considerable attention from the academic and industrial communities. Besides how to construct and train LLMs, how to effectively evaluate and compare the capacity of LLMs has also been well recognized as an important yet difficult problem. Existing paradigms rely on either human annotators or model-based evaluators to evaluate the performance of LLMs on different tasks. However, these paradigms often suffer from high cost, low generalizability, and inherited biases in practice, which make them incapable of supporting the sustainable development of LLMs in long term. In order to address these issues, inspired by the peer review systems widely used in academic publication process, we propose a novel framework that can automatically evaluate LLMs through a peer-review process. Specifically, for the evaluation of a specific task, we first construct a small qualification exam to select "reviewers" from a couple of powerful LLMs. Then, to actually evaluate the "submissions" written by different candidate LLMs, i.e., the evaluatees, we use the reviewer LLMs to rate or compare the submissions. The final ranking of evaluatee LLMs is generated based on the results provided by all reviewers. We conducted extensive experiments on text summarization tasks with eleven LLMs including GPT-4. The results demonstrate the existence of biasness when evaluating using a single LLM. Also, our PRE model outperforms all the baselines, illustrating the effectiveness of the peer review mechanism.

Weihang Su, Xuanyi Chen, Yueyue Wu et al.

Automating the drafting of judgment documents is pivotal to judicial efficiency, yet it remains challenging due to the dual requirements of comprehensive retrieval of legal information and rigorous logical reasoning. Existing approaches, typically relying on standard Retrieval-Augmented Generation and Supervised Fine-Tuning, often suffer from insufficient evidence recall, hallucinated statutory references, and logically flawed legal reasoning. To bridge this gap, we propose Judge-R1, a unified framework designed to enhance LLM-based judgment document generation by jointly improving legal information collection and judgment document generation. First, we introduce Agentic Legal Information Collection, which employs a dynamic planning agent to retrieve precise statutes and precedents from multiple sources. Second, we implement Rubric-Guided Optimization, a reinforcement learning phase utilizing Group Relative Policy Optimization (GRPO) with a comprehensive legal reward function to enforce adherence to judicial standards and reasoning logic. Extensive experiments on the JuDGE benchmark demonstrate that Judge-R1 significantly outperforms state-of-the-art baselines in both legal accuracy and generation quality.

Haitao Li, Qingyao Ai, Jia Chen et al. · tsinghua

Large Language Models (LLMs) like ChatGPT and GPT-4 are versatile and capable of addressing a diverse range of tasks. However, general LLMs, which are developed on open-domain data, may lack the domain-specific knowledge essential for tasks in vertical domains, such as legal, medical, etc. To address this issue, previous approaches either conduct continuous pre-training with domain-specific data or employ retrieval augmentation to support general LLMs. Unfortunately, these strategies are either cost-intensive or unreliable in practical applications. To this end, we present a novel framework named BLADE, which enhances Black-box LArge language models with small Domain-spEcific models. BLADE consists of a black-box LLM and a small domain-specific LM. The small LM preserves domain-specific knowledge and offers specialized insights, while the general LLM contributes robust language comprehension and reasoning capabilities. Specifically, our method involves three steps: 1) pre-training the small LM with domain-specific data, 2) fine-tuning this model using knowledge instruction data, and 3) joint Bayesian optimization of the general LLM and the small LM. Extensive experiments conducted on public legal and medical benchmarks reveal that BLADE significantly outperforms existing approaches. This shows the potential of BLADE as an effective and cost-efficient solution in adapting general LLMs for vertical domains.

Weihang Su, Qingyao Ai, Xiangsheng Li et al.

With the development of deep learning and natural language processing techniques, pre-trained language models have been widely used to solve information retrieval (IR) problems. Benefiting from the pre-training and fine-tuning paradigm, these models achieve state-of-the-art performance. In previous works, plain texts in Wikipedia have been widely used in the pre-training stage. However, the rich structured information in Wikipedia, such as the titles, abstracts, hierarchical heading (multi-level title) structure, relationship between articles, references, hyperlink structures, and the writing organizations, has not been fully explored. In this paper, we devise four pre-training objectives tailored for IR tasks based on the structured knowledge of Wikipedia. Compared to existing pre-training methods, our approach can better capture the semantic knowledge in the training corpus by leveraging the human-edited structured data from Wikipedia. Experimental results on multiple IR benchmark datasets show the superior performance of our model in both zero-shot and fine-tuning settings compared to existing strong retrieval baselines. Besides, experimental results in biomedical and legal domains demonstrate that our approach achieves better performance in vertical domains compared to previous models, especially in scenarios where long text similarity matching is needed.

Haoming Gong, Qingyao Ai, Zhihao Tao et al.

In web search and recommendation systems, user clicks are widely used to train ranking models. However, click data is heavily biased, i.e., users tend to click higher-ranked items (position bias), choose only what was shown to them (selection bias), and trust top results more (trust bias). Without explicitly modeling these biases, the true relevance of ranked items cannot be correctly learned from clicks. Existing Unbiased Learning-to-Rank (ULTR) methods mainly correct position bias and rely on propensity estimation, but they cannot measure remaining bias, provide risk guarantees, or jointly handle multiple bias sources. To overcome these challenges, this paper introduces a novel causal learning-based ranking framework that extends ULTR by combining Structural Causal Models (SCMs) with information-theoretic tools. SCMs specify how clicks are generated and help identify the true relevance signal from click data, while conditional mutual information, measures how much bias leaks into the learned relevance estimates. We use this leakage measure to define a rigorous notion of disentanglement and include it as a regularizer during model training to reduce bias. In addition, we incorporate a causal inference estimator, i.e., doubly robust estimator, to ensure more reliable risk estimation. Experiments on standard Learning-to-Rank benchmarks show that our method consistently reduces measured bias leakage and improves ranking performance, especially in realistic scenarios where multiple biases-such as position and trust bias-interact strongly.

Haitao Li, Junjie Chen, Qingyao Ai et al.

The use of large language models (LLMs) as automated evaluation tools to assess the quality of generated natural language, known as LLMs-as-Judges, has demonstrated promising capabilities and is rapidly gaining widespread attention. However, when applied to pairwise comparisons of candidate responses, LLM-based evaluators often exhibit selection bias. Specifically, their judgments may become inconsistent when the option positions or ID tokens are swapped, compromising the effectiveness and fairness of the evaluation result. To address this challenge, we introduce CalibraEval, a novel label-free method for mitigating selection bias during inference. Specifically, CalibraEval reformulates debiasing as an optimization task aimed at adjusting observed prediction distributions to align with unbiased prediction distributions. To solve this optimization problem, we propose a non-parametric order-preserving algorithm (NOA). This algorithm leverages the partial order relationships between model prediction distributions, thereby eliminating the need for explicit labels and precise mathematical function modeling.Empirical evaluations of LLMs in multiple representative benchmarks demonstrate that CalibraEval effectively mitigates selection bias and improves performance compared to existing debiasing methods. This work marks a step toward building more robust and unbiased automated evaluation frameworks, paving the way for improved reliability in AI-driven assessments

Ziyi Ye, Xiaohui Xie, Qingyao Ai et al.

The Relevance Feedback (RF) process relies on accurate and real-time relevance estimation of feedback documents to improve retrieval performance. Since collecting explicit relevance annotations imposes an extra burden on the user, extensive studies have explored using pseudo-relevance signals and implicit feedback signals as substitutes. However, such signals are indirect indicators of relevance and suffer from complex search scenarios where user interactions are absent or biased. Recently, the advances in portable and high-precision brain-computer interface (BCI) devices have shown the possibility to monitor user's brain activities during search process. Brain signals can directly reflect user's psychological responses to search results and thus it can act as additional and unbiased RF signals. To explore the effectiveness of brain signals in the context of RF, we propose a novel RF framework that combines BCI-based relevance feedback with pseudo-relevance signals and implicit signals to improve the performance of document re-ranking. The experimental results on the user study dataset show that incorporating brain signals leads to significant performance improvement in our RF framework. Besides, we observe that brain signals perform particularly well in several hard search scenarios, especially when implicit signals as feedback are missing or noisy. This reveals when and how to exploit brain signals in the context of RF.

Weihang Su, Hanwen Zhang, Qingyao Ai et al.

Parametric Retrieval-Augmented Generation (PRAG) encodes external documents into lightweight parameter modules that can be retrieved and merged at inference time, offering a promising alternative to in-context retrieval augmentation. Despite its potential, many PRAG implementations train document adapters with task-supervised objectives, which may cause each adapter to encode both document-specific facts and reusable task-solving behavior. This entanglement may make adapter composition less reliable: when multiple adapters are merged at inference time, their overlapping task behaviors can accumulate together with document-specific updates, potentially making the merged adapter less stable and less focused on the intended document knowledge. To examine this issue, we explore Orthogonal Subspace Decomposition (OSD), an adapter-training setup that separates reusable task behavior from document-specific knowledge adapters. Concretely, we first train a Task LoRA to capture reusable task behavior, and then train document LoRAs to encode document-specific knowledge in a orthogonal subspace. This setup provides a controlled way to examine how orthogonalizing task and document LoRA updates affects adapter composition in multi-document PRAG. Experiments across multiple knowledge-intensive tasks and model scales suggest that this orthogonalization strategy can improve compositional robustness in parametric RAG, especially when multiple document adapters are merged.

Yiteng Tu, Weihang Su, Yujia Zhou et al.

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by integrating external knowledge retrieved from a knowledge base. However, its effectiveness is fundamentally constrained by the reliability of both the retriever and the knowledge base. In real-world scenarios, imperfections in these components often lead to the retrieval of noisy, irrelevant, or misleading counterfactual information, ultimately undermining the trustworthiness of RAG systems. To address this challenge, we propose Robust Fine-Tuning (RbFT), a method designed to enhance the resilience of LLMs against retrieval defects through two targeted fine-tuning tasks. Experimental results demonstrate that RbFT significantly improves the robustness of RAG systems across diverse retrieval conditions, surpassing existing methods while maintaining high inference efficiency and compatibility with other robustness techniques.