Jian-Yun Nie

Shitao Xiao, Zheng Liu, Peitian Zhang et al.

We introduce C-Pack, a package of resources that significantly advance the field of general Chinese embeddings. C-Pack includes three critical resources. 1) C-MTEB is a comprehensive benchmark for Chinese text embeddings covering 6 tasks and 35 datasets. 2) C-MTP is a massive text embedding dataset curated from labeled and unlabeled Chinese corpora for training embedding models. 3) C-TEM is a family of embedding models covering multiple sizes. Our models outperform all prior Chinese text embeddings on C-MTEB by up to +10% upon the time of the release. We also integrate and optimize the entire suite of training methods for C-TEM. Along with our resources on general Chinese embedding, we release our data and models for English text embeddings. The English models achieve state-of-the-art performance on MTEB benchmark; meanwhile, our released English data is 2 times larger than the Chinese data. All these resources are made publicly available at https://github.com/FlagOpen/FlagEmbedding.

Tianyi Tang, Junyi Li, Zhipeng Chen et al. · pku

To facilitate research on text generation, this paper presents a comprehensive and unified library, TextBox 2.0, focusing on the use of pre-trained language models (PLMs). To be comprehensive, our library covers $13$ common text generation tasks and their corresponding $83$ datasets and further incorporates $45$ PLMs covering general, translation, Chinese, dialogue, controllable, distilled, prompting, and lightweight PLMs. We also implement $4$ efficient training strategies and provide $4$ generation objectives for pre-training new PLMs from scratch. To be unified, we design the interfaces to support the entire research pipeline (from data loading to training and evaluation), ensuring that each step can be fulfilled in a unified way. Despite the rich functionality, it is easy to use our library, either through the friendly Python API or command line. To validate the effectiveness of our library, we conduct extensive experiments and exemplify four types of research scenarios. The project is released at the link: https://github.com/RUCAIBox/TextBox.

Junyi Li, Tianyi Tang, Jian-Yun Nie et al.

Pretrained language models (PLMs) have made remarkable progress in text generation tasks via fine-tuning. While, it is challenging to fine-tune PLMs in a data-scarce situation. Therefore, it is non-trivial to develop a general and lightweight model that can adapt to various text generation tasks based on PLMs. To fulfill this purpose, the recent prompt-based learning offers a potential solution. In this paper, we improve this technique and propose a novel prompt-based method (PTG) for text generation in a transferable setting. First, PTG learns a set of source prompts for various source generation tasks and then transfers these prompts as target prompts to perform target generation tasks. To consider both task- and instance-level information, we design an adaptive attention mechanism to derive the target prompts. For each data instance, PTG learns a specific target prompt by attending to highly relevant source prompts. In extensive experiments, PTG yields competitive or better results than fine-tuning methods. We release our source prompts as an open resource, where users can add or reuse them to improve new text generation tasks for future research. Code and data can be available at https://github.com/RUCAIBox/Transfer-Prompts-for-Text-Generation.

Lu Yi, Runlin Lei, Liuyi Yao et al.

LLM agents increasingly face long-horizon tasks such as web search and deep research in real-world applications, where accumulated context can cause long-context degradation and reasoning failures. Prior work mitigates this through context management with agent-side context control or fixed strategies such as summarization, which require training the agent itself for adaptation - making it impractical for closed-source agents and ignoring that different agents may require different strategies. We introduce Adaptive Context Management (AdaCoM), which trains an external LLM to manage the context of a frozen agent through flexible modification actions and end-to-end reinforcement learning. Across diverse agents on web search and deep research benchmarks, AdaCoM substantially improves performance by preserving task constraints and progress while pruning stale content. The learned strategies reveal a Fidelity-Reliability Trade-off: agents with higher vanilla ReAct performance benefit from higher-fidelity context preservation, whereas lower-performing agents require more aggressive compression to stay within a reliable reasoning regime. Transfer experiments show that AdaCoM generalizes most effectively across agents with similar capability (measured by vanilla ReAct performance), suggesting a practical path toward reusable context managers for agent systems.

Wayne Xin Zhao, Kun Zhou, Junyi Li et al.

Language is essentially a complex, intricate system of human expressions governed by grammatical rules. It poses a significant challenge to develop capable AI algorithms for comprehending and grasping a language. As a major approach, language modeling has been widely studied for language understanding and generation in the past two decades, evolving from statistical language models to neural language models. Recently, pre-trained language models (PLMs) have been proposed by pre-training Transformer models over large-scale corpora, showing strong capabilities in solving various NLP tasks. Since researchers have found that model scaling can lead to performance improvement, they further study the scaling effect by increasing the model size to an even larger size. Interestingly, when the parameter scale exceeds a certain level, these enlarged language models not only achieve a significant performance improvement but also show some special abilities that are not present in small-scale language models. To discriminate the difference in parameter scale, the research community has coined the term large language models (LLM) for the PLMs of significant size. Recently, the research on LLMs has been largely advanced by both academia and industry, and a remarkable progress is the launch of ChatGPT, which has attracted widespread attention from society. The technical evolution of LLMs has been making an important impact on the entire AI community, which would revolutionize the way how we develop and use AI algorithms. In this survey, we review the recent advances of LLMs by introducing the background, key findings, and mainstream techniques. In particular, we focus on four major aspects of LLMs, namely pre-training, adaptation tuning, utilization, and capacity evaluation. Besides, we also summarize the available resources for developing LLMs and discuss the remaining issues for future directions.

Wayne Xin Zhao, Kun Zhou, Junyi Li et al.

Language is essentially a complex, intricate system of human expressions governed by grammatical rules. It poses a significant challenge to develop capable AI algorithms for comprehending and grasping a language. As a major approach, language modeling has been widely studied for language understanding and generation in the past two decades, evolving from statistical language models to neural language models. Recently, pre-trained language models (PLMs) have been proposed by pre-training Transformer models over large-scale corpora, showing strong capabilities in solving various NLP tasks. Since researchers have found that model scaling can lead to performance improvement, they further study the scaling effect by increasing the model size to an even larger size. Interestingly, when the parameter scale exceeds a certain level, these enlarged language models not only achieve a significant performance improvement but also show some special abilities that are not present in small-scale language models. To discriminate the difference in parameter scale, the research community has coined the term large language models (LLM) for the PLMs of significant size. Recently, the research on LLMs has been largely advanced by both academia and industry, and a remarkable progress is the launch of ChatGPT, which has attracted widespread attention from society. The technical evolution of LLMs has been making an important impact on the entire AI community, which would revolutionize the way how we develop and use AI algorithms. In this survey, we review the recent advances of LLMs by introducing the background, key findings, and mainstream techniques. In particular, we focus on four major aspects of LLMs, namely pre-training, adaptation tuning, utilization, and capacity evaluation. Besides, we also summarize the available resources for developing LLMs and discuss the remaining issues for future directions.

Huimin Chen, Fengran Mo, Yanhao Wang et al.

As privacy issues are receiving increasing attention within the Natural Language Processing (NLP) community, numerous methods have been proposed to sanitize texts subject to differential privacy. However, the state-of-the-art text sanitization mechanisms based on metric local differential privacy (MLDP) do not apply to non-metric semantic similarity measures and cannot achieve good trade-offs between privacy and utility. To address the above limitations, we propose a novel Customized Text (CusText) sanitization mechanism based on the original $ε$-differential privacy (DP) definition, which is compatible with any similarity measure. Furthermore, CusText assigns each input token a customized output set of tokens to provide more advanced privacy protection at the token level. Extensive experiments on several benchmark datasets show that CusText achieves a better trade-off between privacy and utility than existing mechanisms. The code is available at https://github.com/sai4july/CusText.

Le Zhang, Yihong Wu, Fengran Mo et al. · mila

Multi-modal open-domain question answering typically requires evidence retrieval from databases across diverse modalities, such as images, tables, passages, etc. Even Large Language Models (LLMs) like GPT-4 fall short in this task. To enable LLMs to tackle the task in a zero-shot manner, we introduce MoqaGPT, a straightforward and flexible framework. Using a divide-and-conquer strategy that bypasses intricate multi-modality ranking, our framework can accommodate new modalities and seamlessly transition to new models for the task. Built upon LLMs, MoqaGPT retrieves and extracts answers from each modality separately, then fuses this multi-modal information using LLMs to produce a final answer. Our methodology boosts performance on the MMCoQA dataset, improving F1 by +37.91 points and EM by +34.07 points over the supervised baseline. On the MultiModalQA dataset, MoqaGPT surpasses the zero-shot baseline, improving F1 by 9.5 points and EM by 10.1 points, and significantly closes the gap with supervised methods. Our codebase is available at https://github.com/lezhang7/MOQAGPT.

Jimin Huang, Mengxi Xiao, Dong Li et al.

Financial LLMs hold promise for advancing financial tasks and domain-specific applications. However, they are limited by scarce corpora, weak multimodal capabilities, and narrow evaluations, making them less suited for real-world application. To address this, we introduce \textit{Open-FinLLMs}, the first open-source multimodal financial LLMs designed to handle diverse tasks across text, tabular, time-series, and chart data, excelling in zero-shot, few-shot, and fine-tuning settings. The suite includes FinLLaMA, pre-trained on a comprehensive 52-billion-token corpus; FinLLaMA-Instruct, fine-tuned with 573K financial instructions; and FinLLaVA, enhanced with 1.43M multimodal tuning pairs for strong cross-modal reasoning. We comprehensively evaluate Open-FinLLMs across 14 financial tasks, 30 datasets, and 4 multimodal tasks in zero-shot, few-shot, and supervised fine-tuning settings, introducing two new multimodal evaluation datasets. Our results show that Open-FinLLMs outperforms afvanced financial and general LLMs such as GPT-4, across financial NLP, decision-making, and multi-modal tasks, highlighting their potential to tackle real-world challenges. To foster innovation and collaboration across academia and industry, we release all codes (https://anonymous.4open.science/r/PIXIU2-0D70/B1D7/LICENSE) and models under OSI-approved licenses.

Yutao Zhu, Jian-Yun Nie, Yixuan Su et al. · cambridge

Contextual information in search sessions is important for capturing users' search intents. Various approaches have been proposed to model user behavior sequences to improve document ranking in a session. Typically, training samples of (search context, document) pairs are sampled randomly in each training epoch. In reality, the difficulty to understand user's search intent and to judge document's relevance varies greatly from one search context to another. Mixing up training samples of different difficulties may confuse the model's optimization process. In this work, we propose a curriculum learning framework for context-aware document ranking, in which the ranking model learns matching signals between the search context and the candidate document in an easy-to-hard manner. In so doing, we aim to guide the model gradually toward a global optimum. To leverage both positive and negative examples, two curricula are designed. Experiments on two real query log datasets show that our proposed framework can improve the performance of several existing methods significantly, demonstrating the effectiveness of curriculum learning for context-aware document ranking.

Hongjing Qian, Yutao Zhu, Zhicheng Dou et al.

In this paper, we introduce a new NLP task -- generating short factual articles with references for queries by mining supporting evidence from the Web. In this task, called WebBrain, the ultimate goal is to generate a fluent, informative, and factually-correct short article (e.g., a Wikipedia article) for a factual query unseen in Wikipedia. To enable experiments on WebBrain, we construct a large-scale dataset WebBrain-Raw by extracting English Wikipedia articles and their crawlable Wikipedia references. WebBrain-Raw is ten times larger than the previous biggest peer dataset, which can greatly benefit the research community. From WebBrain-Raw, we construct two task-specific datasets: WebBrain-R and WebBrain-G, which are used to train in-domain retriever and generator, respectively. Besides, we empirically analyze the performances of the current state-of-the-art NLP techniques on WebBrain and introduce a new framework ReGen, which enhances the generation factualness by improved evidence retrieval and task-specific pre-training for generation. Experiment results show that ReGen outperforms all baselines in both automatic and human evaluations.

Junyi Li, Tianyi Tang, Wayne Xin Zhao et al.

We study the text generation task under the approach of pre-trained language models (PLMs). Typically, an auto-regressive (AR) method is adopted for generating texts in a token-by-token manner. Despite many advantages of AR generation, it usually suffers from inefficient inference. Therefore, non-autoregressive (NAR) models are proposed to generate all target tokens simultaneously. However, NAR models usually generate texts of lower quality due to the absence of token dependency in the output text. In this paper, we propose ELMER: an efficient and effective PLM for NAR text generation to explicitly model the token dependency during NAR generation. By leveraging the early exit technique, ELMER enables the token generations at different layers, according to their prediction confidence (a more confident token will exit at a lower layer). Besides, we propose a novel pre-training objective, Layer Permutation Language Modeling, to pre-train ELMER by permuting the exit layer for each token in sequences. Experiments on three text generation tasks show that ELMER significantly outperforms NAR models and further narrows the performance gap with AR PLMs (\eg ELMER (29.92) vs BART (30.61) ROUGE-L in XSUM) while achieving over 10 times inference speedup.

Pan Du, Jian-Yun Nie, Yutao Zhu et al.

Beyond topical relevance, passage ranking for open-domain factoid question answering also requires a passage to contain an answer (answerability). While a few recent studies have incorporated some reading capability into a ranker to account for answerability, the ranker is still hindered by the noisy nature of the training data typically available in this area, which considers any passage containing an answer entity as a positive sample. However, the answer entity in a passage is not necessarily mentioned in relation with the given question. To address the problem, we propose an approach called \ttt{PReGAN} for Passage Reranking based on Generative Adversarial Neural networks, which incorporates a discriminator on answerability, in addition to a discriminator on topical relevance. The goal is to force the generator to rank higher a passage that is topically relevant and contains an answer. Experiments on five public datasets show that \ttt{PReGAN} can better rank appropriate passages, which in turn, boosts the effectiveness of QA systems, and outperforms the existing approaches without using external data.

Fengran Mo, Jian-Yun Nie, Kaiyu Huang et al.

Conversational search allows a user to interact with a search system in multiple turns. A query is strongly dependent on the conversation context. An effective way to improve retrieval effectiveness is to expand the current query with historical queries. However, not all the previous queries are related to, and useful for expanding the current query. In this paper, we propose a new method to select relevant historical queries that are useful for the current query. To cope with the lack of labeled training data, we use a pseudo-labeling approach to annotate useful historical queries based on their impact on the retrieval results. The pseudo-labeled data are used to train a selection model. We further propose a multi-task learning framework to jointly train the selector and the retriever during fine-tuning, allowing us to mitigate the possible inconsistency between the pseudo labels and the changed retriever. Extensive experiments on four conversational search datasets demonstrate the effectiveness and broad applicability of our method compared with several strong baselines.

Junyi Li, Wayne Xin Zhao, Jian-Yun Nie et al.

Diffusion models have become a new generative paradigm for text generation. Considering the discrete categorical nature of text, in this paper, we propose GlyphDiffusion, a novel diffusion approach for text generation via text-guided image generation. Our key idea is to render the target text as a glyph image containing visual language content. In this way, conditional text generation can be cast as a glyph image generation task, and it is then natural to apply continuous diffusion models to discrete texts. Specially, we utilize a cascaded architecture (ie a base and a super-resolution diffusion model) to generate high-fidelity glyph images, conditioned on the input text. Furthermore, we design a text grounding module to transform and refine the visual language content from generated glyph images into the final texts. In experiments over four conditional text generation tasks and two classes of metrics (ie quality and diversity), GlyphDiffusion can achieve comparable or even better results than several baselines, including pretrained language models. Our model also makes significant improvements compared to the recent diffusion model.

Rui Qi, Fengran Mo, Yufeng Chen et al.

Multilingual retrieval-augmented generation (MRAG) requires models to effectively acquire and integrate beneficial external knowledge from multilingual collections. However, most existing studies employ a unitive process where queries of equivalent semantics across different languages are processed through a single-turn retrieval and subsequent optimization. Such a ``one-size-fits-all'' strategy is often suboptimal in multilingual settings, as the models occur to knowledge bias and conflict during the interaction with the search engine. To alleviate the issues, we propose LcRL, a multilingual search-augmented reinforcement learning framework that integrates a language-coupled Group Relative Policy Optimization into the policy and reward models. We adopt the language-coupled group sampling in the rollout module to reduce knowledge bias, and regularize an auxiliary anti-consistency penalty in the reward models to mitigate the knowledge conflict. Experimental results demonstrate that LcRL not only achieves competitive performance but is also appropriate for various practical scenarios such as constrained training data and retrieval over collections encompassing a large number of languages. Our code is available at https://github.com/Cherry-qwq/LcRL-Open.

Yihan Wu, Xi Wang, Shaofei Zhang et al.

Expressive speech synthesis, like audiobook synthesis, is still challenging for style representation learning and prediction. Deriving from reference audio or predicting style tags from text requires a huge amount of labeled data, which is costly to acquire and difficult to define and annotate accurately. In this paper, we propose a novel framework for learning style representation from abundant plain text in a self-supervised manner. It leverages an emotion lexicon and uses contrastive learning and deep clustering. We further integrate the style representation as a conditioned embedding in a multi-style Transformer TTS. Comparing with multi-style TTS by predicting style tags trained on the same dataset but with human annotations, our method achieves improved results according to subjective evaluations on both in-domain and out-of-domain test sets in audiobook speech. Moreover, with implicit context-aware style representation, the emotion transition of synthesized audio in a long paragraph appears more natural. The audio samples are available on the demo web.

Fengran Mo, Chen Qu, Kelong Mao et al.

Conversational search supports multi-turn user-system interactions to solve complex information needs. Different from the traditional single-turn ad-hoc search, conversational search encounters a more challenging problem of context-dependent query understanding with the lengthy and long-tail conversational history context. While conversational query rewriting methods leverage explicit rewritten queries to train a rewriting model to transform the context-dependent query into a stand-stone search query, this is usually done without considering the quality of search results. Conversational dense retrieval methods use fine-tuning to improve a pre-trained ad-hoc query encoder, but they are limited by the conversational search data available for training. In this paper, we leverage both rewritten queries and relevance judgments in the conversational search data to train a better query representation model. The key idea is to align the query representation with those of rewritten queries and relevant documents. The proposed model -- Query Representation Alignment Conversational Dense Retriever, QRACDR, is tested on eight datasets, including various settings in conversational search and ad-hoc search. The results demonstrate the strong performance of QRACDR compared with state-of-the-art methods, and confirm the effectiveness of representation alignment.

Fengran Mo, Longxiang Zhao, Kaiyu Huang et al.

Personalized conversational information retrieval (CIR) combines conversational and personalizable elements to satisfy various users' complex information needs through multi-turn interaction based on their backgrounds. The key promise is that the personal textual knowledge base (PTKB) can improve the CIR effectiveness because the retrieval results can be more related to the user's background. However, PTKB is noisy: not every piece of knowledge in PTKB is relevant to the specific query at hand. In this paper, we explore and test several ways to select knowledge from PTKB and use it for query reformulation by using a large language model (LLM). The experimental results show the PTKB might not always improve the search results when used alone, but LLM can help generate a more appropriate personalized query when high-quality guidance is provided.

Xinying Qiu, Shuxuan Liao, Jiajun Xie et al.

In the prompt-specific holistic score prediction task for Automatic Essay Scoring, the general approaches include pre-trained neural model, coherence model, and hybrid model that incorporate syntactic features with neural model. In this paper, we propose a novel approach to extract and represent essay coherence features with prompt-learning NSP that shows to match the state-of-the-art AES coherence model, and achieves the best performance for long essays. We apply syntactic feature dense embedding to augment BERT-based model and achieve the best performance for hybrid methodology for AES. In addition, we explore various ideas to combine coherence, syntactic information and semantic embeddings, which no previous study has done before. Our combined model also performs better than the SOTA available for combined model, even though it does not outperform our syntactic enhanced neural model. We further offer analyses that can be useful for future study.

Junyi Li, Jie Chen, Ruiyang Ren et al.

In the era of large language models (LLMs), hallucination (i.e., the tendency to generate factually incorrect content) poses great challenge to trustworthy and reliable deployment of LLMs in real-world applications. To tackle the LLM hallucination, three key questions should be well studied: how to detect hallucinations (detection), why do LLMs hallucinate (source), and what can be done to mitigate them (mitigation). To address these challenges, this work presents a systematic empirical study on LLM hallucination, focused on the the three aspects of hallucination detection, source and mitigation. Specially, we construct a new hallucination benchmark HaluEval 2.0, and designs a simple yet effective detection method for LLM hallucination. Furthermore, we zoom into the different training or utilization stages of LLMs and extensively analyze the potential factors that lead to the LLM hallucination. Finally, we implement and examine a series of widely used techniques to mitigate the hallucinations in LLMs. Our work has led to several important findings to understand the hallucination origin and mitigate the hallucinations in LLMs. Our code and data can be accessed at https://github.com/RUCAIBox/HaluEval-2.0.

Rui Qi, Fengran Mo, Sijin Lu et al.

A multilingual collection may contain useful knowledge in other languages to supplement and correct the facts in the original language for Retrieval-Augmented Generation (RAG). However, the vanilla approach that simply concatenates multiple pieces of knowledge from different languages into the context may fail to improve effectiveness due to the potential disparities across languages. To better leverage multilingual knowledge, we propose CroSearch-R1, a search-augmented reinforcement learning framework to integrate multilingual knowledge into the Group Relative Policy Optimization (GRPO) process. In particular, the approach adopts a multi-turn retrieval strategy with cross-lingual knowledge integration to dynamically align the knowledge from other languages as supplementary evidence into a unified representation space. Furthermore, we introduce a multilingual rollout mechanism to optimize reasoning transferability across languages. Experimental results demonstrate that our framework effectively leverages cross-lingual complementarity and improves the effectiveness of RAG with multilingual collections.

Yan Wang, Yi Han, Lingfei Qian et al.

Most recommendation benchmarks evaluate how well a model imitates user behavior. In financial advisory, however, observed actions can be noisy or short-sighted under market volatility and may conflict with a user's long-term goals. Treating what users chose as the sole ground truth, therefore, conflates behavioral imitation with decision quality. We introduce Conv-FinRe, a conversational and longitudinal benchmark for stock recommendation that evaluates LLMs beyond behavior matching. Given an onboarding interview, step-wise market context, and advisory dialogues, models must generate rankings over a fixed investment horizon. Crucially, Conv-FinRe provides multi-view references that distinguish descriptive behavior from normative utility grounded in investor-specific risk preferences, enabling diagnosis of whether an LLM follows rational analysis, mimics user noise, or is driven by market momentum. We build the benchmark from real market data and human decision trajectories, instantiate controlled advisory conversations, and evaluate a suite of state-of-the-art LLMs. Results reveal a persistent tension between rational decision quality and behavioral alignment: models that perform well on utility-based ranking often fail to match user choices, whereas behaviorally aligned models can overfit short-term noise. The dataset is publicly released on Hugging Face, and the codebase is available on GitHub.

Jiayin Wang, Fengran Mo, Weizhi Ma et al.

Large language models (LLMs) are essential tools that users employ across various scenarios, so evaluating their performance and guiding users in selecting the suitable service is important. Although many benchmarks exist, they mainly focus on specific predefined model abilities, such as world knowledge, reasoning, etc. Based on these ability scores, it is hard for users to determine which LLM best suits their particular needs. To address these issues, we propose to evaluate LLMs from a user-centric perspective and design this benchmark to measure their efficacy in satisfying user needs under distinct intents. Firstly, we collect 1,846 real-world use cases from a user study with 712 participants from 23 countries. This first-hand data helps us understand actual user intents and needs in LLM interactions, forming the User Reported Scenarios (URS) dataset, which is categorized with six types of user intents. Secondly, based on this authentic dataset, we benchmark 10 LLM services with GPT-4-as-Judge. Thirdly, we show that benchmark scores align well with human preference in both real-world experience and pair-wise annotations, achieving Pearson correlations of 0.95 and 0.94, respectively. This alignment confirms that the URS dataset and our evaluation method establish an effective user-centric benchmark. The dataset, code, and process data are available at https://github.com/Alice1998/URS.

Yuxing Tian, Yiyan Qi, Fengran Mo et al.

Dynamic graph anomaly detection (DGAD) is critical for many real-world applications but remains challenging due to the scarcity of labeled anomalies. Existing methods are either unsupervised or semi-supervised: unsupervised methods avoid the need for labeled anomalies but often produce ambiguous boundary, whereas semi-supervised methods can overfit to the limited labeled anomalies and generalize poorly to unseen anomalies. To address this gap, we consider a largely underexplored problem in DGAD: learning a discriminative boundary from normal/unlabeled data, while leveraging limited labeled anomalies \textbf{when available} without sacrificing generalization to unseen anomalies. To this end, we propose an effective, generalizable, and model-agnostic framework with three main components: (i) residual representation encoding that capture deviations between current interactions and their historical context, providing anomaly-relevant signals; (ii) a restriction loss that constrain the normal representations within an interval bounded by two co-centered hyperspheres, ensuring consistent scales while keeping anomalies separable; (iii) a bi-boundary optimization strategy that learns a discriminative and robust boundary using the normal log-likelihood distribution modeled by a normalizing flow. Extensive experiments demonstrate the superiority of our framework across diverse evaluation settings.

Yuxing Tian, Fengran Mo, Weixu Zhang et al.

The strong capabilities of recent Large Language Models (LLMs) have made them highly effective for zero-shot re-ranking task. Attention-based re-ranking methods, which derive relevance scores directly from attention weights, offer an efficient and interpretable alternative to generation-based re-ranking methods. However, they still face two major limitations. First, attention signals are highly concentrated a small subset of tokens within a few documents, making others indistinguishable. Second, attention often overemphasizes phrases lexically similar to the query, yielding biased rankings that irrelevant documents with mere lexical resemblance are regarded as relevant. In this paper, we propose \textbf{ReAttn}, a post-hoc re-weighting strategy for attention-based re-ranking methods. It first compute the cross-document IDF weighting to down-weight attention on query-overlapping tokens that frequently appear across the candidate documents, reducing lexical bias and emphasizing distinctive terms. It then employs entropy-based regularization to mitigate over-concentrated attention, encouraging a more balanced distribution across informative tokens. Both adjustments operate directly on existing attention weights without additional training or supervision. Extensive experiments demonstrate the effectiveness of our method.

Qiwei Shao, Fengran Mo, Jian-Yun Nie

Document-level biomedical concept extraction is the task of identifying biomedical concepts mentioned in a given document. Recent advancements have adapted pre-trained language models for this task. However, the scarcity of domain-specific data and the deviation of concepts from their canonical names often hinder these models' effectiveness. To tackle this issue, we employ MetaMapLite, an existing rule-based concept mapping system, to generate additional pseudo-annotated data from PubMed and PMC. The annotated data are used to augment the limited training data. Through extensive experiments, this study demonstrates the utility of a manually crafted concept mapping tool for training a better concept extraction model.

Yi Han, Lingfei Qian, Yan Wang et al.

Large language models (LLMs) have enabled agentic systems that can reason, plan, and act across complex tasks, but it remains unclear whether they can allocate resources effectively under uncertainty. Unlike short-horizon reactive decisions, allocation requires committing scarce resources over time while balancing competing objectives and preserving flexibility for future needs. We introduce EnterpriseArena, the first benchmark for evaluating agents on long-horizon enterprise resource allocation. It instantiates CFO-style decision-making in a 132-month enterprise simulator combining firm-level financial data, anonymized business documents, macroeconomic and industry signals, and expert-validated operating rules. The environment is partially observable and reveals the state only through budgeted organizational tools, forcing agents to trade off information acquisition against conserving scarce resources. Experiments on eleven advanced LLMs show that this setting remains highly challenging: only 16% of runs survive the full horizon, and larger models do not reliably outperform smaller ones. These results identify long-horizon resource allocation under uncertainty as a distinct capability gap for current LLM agents.

Long Yuan, Fengran Mo, Kaiyu Huang et al.

The rapid advancement of multimodal large language models (LLMs) has opened new frontiers in artificial intelligence, enabling the integration of diverse large-scale data types such as text, images, and spatial information. In this paper, we explore the potential of multimodal LLMs (MLLM) for geospatial artificial intelligence (GeoAI), a field that leverages spatial data to address challenges in domains including Geospatial Semantics, Health Geography, Urban Geography, Urban Perception, and Remote Sensing. We propose a MLLM (OmniGeo) tailored to geospatial applications, capable of processing and analyzing heterogeneous data sources, including satellite imagery, geospatial metadata, and textual descriptions. By combining the strengths of natural language understanding and spatial reasoning, our model enhances the ability of instruction following and the accuracy of GeoAI systems. Results demonstrate that our model outperforms task-specific models and existing LLMs on diverse geospatial tasks, effectively addressing the multimodality nature while achieving competitive results on the zero-shot geospatial tasks. Our code will be released after publication.

Zhan Su, Fengran Mo, Jinghan Zhang et al.

The \textit{de facto} paradigm for applying dense retrieval (DR) to new tasks involves fine-tuning a pre-trained model for a specific task. However, this paradigm has two significant limitations: (1) It is difficult adapt the DR to a new domain if the training dataset is limited. (2) Old DR models are simply replaced by newer models that are trained from scratch when the former are no longer up to date. Especially for scenarios where the model needs to be updated frequently, this paradigm is prohibitively expensive. To address these challenges, we propose a novel dense retrieval approach, termed \textit{dynamic dense retrieval} (DDR). DDR uses \textit{prefix tuning} as a \textit{module} specialized for a specific domain. These modules can then be compositional combined with a dynamic routing strategy, enabling highly flexible domain adaptation in the retrieval part. Extensive evaluation on six zero-shot downstream tasks demonstrates that this approach can surpass DR while utilizing only 2\% of the training parameters, paving the way to achieve more flexible dense retrieval in IR. We see it as a promising future direction for applying dense retrieval to various tasks.

Xueqing Peng, Zhuohan Xie, Yupeng Cao et al.

As AI agents improve, the central question is no longer whether they can solve isolated well-defined financial tasks, but whether they can reliably carry out financial professional work. Existing financial benchmarks offer only a partial view of this ability, as they primarily evaluate static competencies such as question answering, retrieval, summarization, and classification. We introduce Herculean, the first skilled benchmark for agentic financial intelligence spanning four representative workflows, including Trading, Hedging, Market Insights, and Auditing. Each workflow is instantiated as a standardized MCP-based skill environment with its own tools, interaction dynamics, constraints, and success criteria, enabling consistent end-to-end assessment of heterogeneous agent systems. Across frontier agents, we find agents perform relatively well on Trading and Market Insights, but struggle substantially on Hedging and Auditing, where long-horizon coordination, state consistency, and structured verification are critical. Overall, our results point to a key gap in current agents in turning financial reasoning into dependable workflow execution in high-stakes financial workflows.

Shuo Zhang, Fabrizio Gotti, Fengran Mo et al.

Hallucination in large language models (LLMs) is a fundamental challenge, particularly in open-domain question answering. Prior work attempts to detect hallucination with model-internal signals such as token-level entropy or generation consistency, while the connection between pretraining data exposure and hallucination is underexplored. Existing studies show that LLMs underperform on long-tail knowledge, i.e., the accuracy of the generated answer drops for the ground-truth entities that are rare in pretraining. However, examining whether data coverage itself can serve as a detection signal is overlooked. We propose a complementary question: Does lexical training-data coverage of the question and/or generated answer provide additional signal for hallucination detection? To investigate this, we construct scalable suffix arrays over RedPajama's 1.3-trillion-token pretraining corpus to retrieve $n$-gram statistics for both prompts and model generations. We evaluate their effectiveness for hallucination detection across three QA benchmarks. Our observations show that while occurrence-based features are weak predictors when used alone, they yield modest gains when combined with log-probabilities, particularly on datasets with higher intrinsic model uncertainty. These findings suggest that lexical coverage features provide a complementary signal for hallucination detection. All code and suffix-array infrastructure are provided at https://github.com/WWWonderer/ostd.

Fengran Mo, Abbas Ghaddar, Kelong Mao et al.

In this paper, we study how open-source large language models (LLMs) can be effectively deployed for improving query rewriting in conversational search, especially for ambiguous queries. We introduce CHIQ, a two-step method that leverages the capabilities of LLMs to resolve ambiguities in the conversation history before query rewriting. This approach contrasts with prior studies that predominantly use closed-source LLMs to directly generate search queries from conversation history. We demonstrate on five well-established benchmarks that CHIQ leads to state-of-the-art results across most settings, showing highly competitive performances with systems leveraging closed-source LLMs. Our study provides a first step towards leveraging open-source LLMs in conversational search, as a competitive alternative to the prevailing reliance on commercial LLMs. Data, models, and source code will be publicly available upon acceptance at https://github.com/fengranMark/CHIQ.

Junyi Li, Xiaoxue Cheng, Wayne Xin Zhao et al.

Large language models (LLMs), such as ChatGPT, are prone to generate hallucinations, i.e., content that conflicts with the source or cannot be verified by the factual knowledge. To understand what types of content and to which extent LLMs are apt to hallucinate, we introduce the Hallucination Evaluation benchmark for Large Language Models (HaluEval), a large collection of generated and human-annotated hallucinated samples for evaluating the performance of LLMs in recognizing hallucination. To generate these samples, we propose a ChatGPT-based two-step framework, i.e., sampling-then-filtering. Besides, we also hire some human labelers to annotate the hallucinations in ChatGPT responses. The empirical results suggest that ChatGPT is likely to generate hallucinated content in specific topics by fabricating unverifiable information (i.e., about $19.5\%$ responses). Moreover, existing LLMs face great challenges in recognizing the hallucinations in texts. However, our experiments also prove that providing external knowledge or adding reasoning steps can help LLMs recognize hallucinations. Our benchmark can be accessed at https://github.com/RUCAIBox/HaluEval.

Yutao Zhu, Jian-Yun Nie, Kun Zhou et al.

A proactive dialogue system has the ability to proactively lead the conversation. Different from the general chatbots which only react to the user, proactive dialogue systems can be used to achieve some goals, e.g., to recommend some items to the user. Background knowledge is essential to enable smooth and natural transitions in dialogue. In this paper, we propose a new multi-task learning framework for retrieval-based knowledge-grounded proactive dialogue. To determine the relevant knowledge to be used, we frame knowledge prediction as a complementary task and use explicit signals to supervise its learning. The final response is selected according to the predicted knowledge, the goal to achieve, and the context. Experimental results show that explicit modeling of knowledge prediction and goal selection can greatly improve the final response selection. Our code is available at https://github.com/DaoD/KPN/.

Yutao Zhu, Kun Zhou, Jian-Yun Nie et al.

Sentence ordering aims at arranging a list of sentences in the correct order. Based on the observation that sentence order at different distances may rely on different types of information, we devise a new approach based on multi-granular orders between sentences. These orders form multiple constraint graphs, which are then encoded by Graph Isomorphism Networks and fused into sentence representations. Finally, sentence order is determined using the order-enhanced sentence representations. Our experiments on five benchmark datasets show that our method outperforms all the existing baselines significantly, achieving a new state-of-the-art performance. The results demonstrate the advantage of considering multiple types of order information and using graph neural networks to integrate sentence content and order information for the task. Our code is available at https://github.com/DaoD/ConstraintGraph4NSO.

Yutao Zhu, Jian-Yun Nie, Kun Zhou et al.

Grounding human-machine conversation in a document is an effective way to improve the performance of retrieval-based chatbots. However, only a part of the document content may be relevant to help select the appropriate response at a round. It is thus crucial to select the part of document content relevant to the current conversation context. In this paper, we propose a document content selection network (CSN) to perform explicit selection of relevant document contents, and filter out the irrelevant parts. We show in experiments on two public document-grounded conversation datasets that CSN can effectively help select the relevant document contents to the conversation context, and it produces better results than the state-of-the-art approaches. Our code and datasets are available at https://github.com/DaoD/CSN.

Wenhao Xu, Akshatha Arodi, Jian-Yun Nie et al.

Modern slavery affects millions of people worldwide, and regulatory frameworks such as Modern Slavery Acts now require companies to publish detailed disclosures. However, these statements are often vague and inconsistent, making manual review time-consuming and difficult to scale. While NLP offers a promising path forward, high-stakes compliance tasks require more than accurate classification: they demand transparent, rule-aligned outputs that legal experts can verify. Existing applications of large language models (LLMs) often reduce complex regulatory assessments to binary decisions, lacking the necessary structure for robust legal scrutiny. We argue that compliance verification is fundamentally a rule-matching problem: it requires evaluating whether textual statements adhere to well-defined regulatory rules. To this end, we propose a novel framework that harnesses AI for rule-level compliance verification while preserving expert oversight. At its core is the Compliance Alignment Judge (CA-Judge), which evaluates model-generated justifications based on their fidelity to statutory requirements. Using this feedback, we train the Compliance Alignment LLM (CALLM), a model that produces rule-consistent, human-verifiable outputs. CALLM improves predictive performance and generates outputs that are both transparent and legally grounded, offering a more verifiable and actionable solution for real-world compliance analysis.

Kaiyu Huang, Fengran Mo, Xinyu Zhang et al. · tsinghua

The rapid development of Large Language Models (LLMs) demonstrates remarkable multilingual capabilities in natural language processing, attracting global attention in both academia and industry. To mitigate potential discrimination and enhance the overall usability and accessibility for diverse language user groups, it is important for the development of language-fair technology. Despite the breakthroughs of LLMs, the investigation into the multilingual scenario remains insufficient, where a comprehensive survey to summarize recent approaches, developments, limitations, and potential solutions is desirable. To this end, we provide a survey with multiple perspectives on the utilization of LLMs in the multilingual scenario. We first rethink the transitions between previous and current research on pre-trained language models. Then we introduce several perspectives on the multilingualism of LLMs, including training and inference methods, information retrieval, model security, multi-domain with language culture, and usage of datasets. We also discuss the major challenges that arise in these aspects, along with possible solutions. Besides, we highlight future research directions that aim at further enhancing LLMs with multilingualism. The survey aims to help the research community address multilingual problems and provide a comprehensive understanding of the core concepts, key techniques, and latest developments in multilingual natural language processing based on LLMs.

Yujia Zhou, Zheng Liu, Jiajie Jin et al.

Retrieval-augmented generation have become central in natural language processing due to their efficacy in generating factual content. While traditional methods employ single-time retrieval, more recent approaches have shifted towards multi-time retrieval for multi-hop reasoning tasks. However, these strategies are bound by predefined reasoning steps, potentially leading to inaccuracies in response generation. This paper introduces MetaRAG, an approach that combines the retrieval-augmented generation process with metacognition. Drawing from cognitive psychology, metacognition allows an entity to self-reflect and critically evaluate its cognitive processes. By integrating this, MetaRAG enables the model to monitor, evaluate, and plan its response strategies, enhancing its introspective reasoning abilities. Through a three-step metacognitive regulation pipeline, the model can identify inadequacies in initial cognitive responses and fixes them. Empirical evaluations show that MetaRAG significantly outperforms existing methods.

Fengran Mo, Kelong Mao, Ziliang Zhao et al.

As a cornerstone of modern information access, search engines have become indispensable in everyday life. With the rapid advancements in AI and natural language processing (NLP) technologies, particularly large language models (LLMs), search engines have evolved to support more intuitive and intelligent interactions between users and systems. Conversational search, an emerging paradigm for next-generation search engines, leverages natural language dialogue to facilitate complex and precise information retrieval, thus attracting significant attention. Unlike traditional keyword-based search engines, conversational search systems enhance user experience by supporting intricate queries, maintaining context over multi-turn interactions, and providing robust information integration and processing capabilities. Key components such as query reformulation, search clarification, conversational retrieval, and response generation work in unison to enable these sophisticated interactions. In this survey, we explore the recent advancements and potential future directions in conversational search, examining the critical modules that constitute a conversational search system. We highlight the integration of LLMs in enhancing these systems and discuss the challenges and opportunities that lie ahead in this dynamic field. Additionally, we provide insights into real-world applications and robust evaluations of current conversational search systems, aiming to guide future research and development in conversational search.

Fengran Mo, Bole Yi, Kelong Mao et al.

Conversational search provides a more convenient interface for users to search by allowing multi-turn interaction with the search engine. However, the effectiveness of the conversational dense retrieval methods is limited by the scarcity of training data required for their fine-tuning. Thus, generating more training conversational sessions with relevant labels could potentially improve search performance. Based on the promising capabilities of large language models (LLMs) on text generation, we propose ConvSDG, a simple yet effective framework to explore the feasibility of boosting conversational search by using LLM for session data generation. Within this framework, we design dialogue/session-level and query-level data generation with unsupervised and semi-supervised learning, according to the availability of relevance judgments. The generated data are used to fine-tune the conversational dense retriever. Extensive experiments on four widely used datasets demonstrate the effectiveness and broad applicability of our ConvSDG framework compared with several strong baselines.

Yuxing Tian, Fengran Mo, Zhiqi Huang et al.

Large Language Models (LLMs) have recently been explored as fine-grained zero-shot re-rankers by leveraging attention signals to estimate document relevance. However, existing methods either aggregate attention signals across all heads or rely on a statically selected subset identified by heuristic rules. This solution can be suboptimal because the informative heads can vary across queries or domains. Moreover, naively combining multiple heads can degrade performance due to redundancy or conflicting ranking signals. In this paper, we propose a query-dependent head selection method, RouteHead, for attention-based re-ranking with LLMs. Specifically, we learn a lightweight router that can map each query to an optimal head set, and relevance scores are computed by aggregating attention signals only from these heads. Since query-to-head optimal labels are unavailable, we first construct pseudo labels via an offline search. The router represents each head with a learnable embedding and represents each query using an embedding extracted from the hidden states of the frozen LLM. Then it is trained on the pseudo labels with a sparsity regularizer. Experiments on diverse benchmarks and multiple LLM backbones show that the proposed method consistently outperforms strong baselines.

Fengran Mo, Chuan Meng, Mohammad Aliannejadi et al.

Conversational search enables multi-turn interactions between users and systems to fulfill users' complex information needs. During this interaction, the system should understand the users' search intent within the conversational context and then return the relevant information through a flexible, dialogue-based interface. The recent powerful large language models (LLMs) with capacities of instruction following, content generation, and reasoning, attract significant attention and advancements, providing new opportunities and challenges for building up intelligent conversational search systems. This tutorial aims to introduce the connection between fundamentals and the emerging topics revolutionized by LLMs in the context of conversational search. It is designed for students, researchers, and practitioners from both academia and industry. Participants will gain a comprehensive understanding of both the core principles and cutting-edge developments driven by LLMs in conversational search, equipping them with the knowledge needed to contribute to the development of next-generation conversational search systems.

Yupeng Cao, Haohang Li, Yangyang Yu et al.

Audio Large Language Models (AudioLLMs) have received widespread attention and have significantly improved performance on audio tasks such as conversation, audio understanding, and automatic speech recognition (ASR). Despite these advancements, there is an absence of a benchmark for assessing AudioLLMs in financial scenarios, where audio data, such as earnings conference calls and CEO speeches, are crucial resources for financial analysis and investment decisions. In this paper, we introduce \textsc{FinAudio}, the first benchmark designed to evaluate the capacity of AudioLLMs in the financial domain. We first define three tasks based on the unique characteristics of the financial domain: 1) ASR for short financial audio, 2) ASR for long financial audio, and 3) summarization of long financial audio. Then, we curate two short and two long audio datasets, respectively, and develop a novel dataset for financial audio summarization, comprising the \textsc{FinAudio} benchmark. Then, we evaluate seven prevalent AudioLLMs on \textsc{FinAudio}. Our evaluation reveals the limitations of existing AudioLLMs in the financial domain and offers insights for improving AudioLLMs. All datasets and codes will be released.

Fengran Mo, Jinghan Zhang, Yuchen Hui et al.

Conversational search aims to satisfy users' complex information needs via multiple-turn interactions. The key challenge lies in revealing real users' search intent from the context-dependent queries. Previous studies achieve conversational search by fine-tuning a conversational dense retriever with relevance judgments between pairs of context-dependent queries and documents. However, this training paradigm encounters data scarcity issues. To this end, we propose ConvMix, a mixed-criteria framework to augment conversational dense retrieval, which covers more aspects than existing data augmentation frameworks. We design a two-sided relevance judgment augmentation schema in a scalable manner via the aid of large language models. Besides, we integrate the framework with quality control mechanisms to obtain semantically diverse samples and near-distribution supervisions to combine various annotated data. Experimental results on five widely used benchmarks show that the conversational dense retriever trained by our ConvMix framework outperforms previous baseline methods, which demonstrates our superior effectiveness.

Yihong Wu, Liheng Ma, Muzhi Li et al.

Large Language Models (LLMs) have demonstrated remarkable versatility, due to the lack of factual knowledge, their application to Question Answering (QA) tasks remains hindered by hallucination. While Retrieval-Augmented Generation mitigates these issues by integrating external knowledge, existing approaches rely heavily on in-context learning, whose performance is constrained by the fundamental reasoning capabilities of LLMs. In this paper, we propose Mujica, a Multi-hop Joint Intelligence for Complex Question Answering, comprising a planner that decomposes questions into a directed acyclic graph of subquestions and a worker that resolves questions via retrieval and reasoning. Additionally, we introduce MyGO (Minimalist policy Gradient Optimization), a novel reinforcement learning method that replaces traditional policy gradient updates with Maximum Likelihood Estimation (MLE) by sampling trajectories from an asymptotically optimal policy. MyGO eliminates the need for gradient rescaling and reference models, ensuring stable and efficient training. Empirical results across multiple datasets demonstrate the effectiveness of Mujica-MyGO in enhancing multi-hop QA performance for various LLMs, offering a scalable and resource-efficient solution for complex QA tasks.

Lingfei Qian, Weipeng Zhou, Yan Wang et al.

As the fundamental capability behind decision-making in finance, financial reasoning poses distinct challenges for LLMs. Although reinforcement learning (RL) have boosted generic reasoning, the progress in finance is hindered by the absence of empirical study of building effective financial chain-of-thought (CoT) corpus, a systematic comparison of different RL methods, and comprehensive benchmarks. To address these gaps, we introduce FinCoT, the first open high-fidelity CoT corpus for finance, distilled from seven QA datasets by a novel three-stage pipeline that incorporates domain supervision, iterative LLM refinement, and difficulty-aware filtering. Based on FinCoT, we develop Fin-o1, the first open financial reasoning models trained via supervised fine-tuning and GRPO-based RL. Our models outperform existing financial reasoning models and SOTA general models such as GPT-o1, DeepSeek-R1, and GPT-4.5. We also investigate the effectiveness of three different RL methods in improving domain-specific reasoning, offering the first such empirical study. We finally propose FinReason, the first financial reasoning benchmark covering multi-table analysis, long-context reasoning, and equation-based tasks, and evaluate 29 LLMs. Our extensive experiments reveal general reasoning models excel on standard benchmarks yet exhibit obvious performance degradation in financial contexts; even finance-tuned models like Dianjin-R1 and FinR1 degrade on lengthy documents. In contrast, our Fin-o1 models consistently outperform their backbones and larger GPT-o1 and DeepSeek-R1, confirming the effectiveness of our data building and model training strategy. Our study further shows that GRPO yields reliable gains whereas PPO and DPO do not, highlighting the need for targeted data and optimisation rather than scale alone.

Pierre Erbacher, Jian-Yun Nie, Philippe Preux et al.

Conversational systems have made significant progress in generating natural language responses. However, their potential as conversational search systems is currently limited due to their passive role in the information-seeking process. One major limitation is the scarcity of datasets that provide labelled ambiguous questions along with a supporting corpus of documents and relevant clarifying questions. This work aims to tackle the challenge of generating relevant clarifying questions by taking into account the inherent ambiguities present in both user queries and documents. To achieve this, we propose PAQA, an extension to the existing AmbiNQ dataset, incorporating clarifying questions. We then evaluate various models and assess how passage retrieval impacts ambiguity detection and the generation of clarifying questions. By addressing this gap in conversational search systems, we aim to provide additional supervision to enhance their active participation in the information-seeking process and provide users with more accurate results.

Fengran Mo, Chen Qu, Kelong Mao et al.

Conversational search facilitates complex information retrieval by enabling multi-turn interactions between users and the system. Supporting such interactions requires a comprehensive understanding of the conversational inputs to formulate a good search query based on historical information. In particular, the search query should include the relevant information from the previous conversation turns. However, current approaches for conversational dense retrieval primarily rely on fine-tuning a pre-trained ad-hoc retriever using the whole conversational search session, which can be lengthy and noisy. Moreover, existing approaches are limited by the amount of manual supervision signals in the existing datasets. To address the aforementioned issues, we propose a History-Aware Conversational Dense Retrieval (HAConvDR) system, which incorporates two ideas: context-denoised query reformulation and automatic mining of supervision signals based on the actual impact of historical turns. Experiments on two public conversational search datasets demonstrate the improved history modeling capability of HAConvDR, in particular for long conversations with topic shifts.