论文检索

Zhiwei Liu, Kailai Yang, Qianqian Xie, Christine de Kock, Sophia Ananiadou, Eduard Hovy

Misinformation is prevalent in various fields such as education, politics, health, etc., causing significant harm to society. However, current methods for cross-domain misinformation detection rely on effort- and resource-intensive fine-tuning and complex model structures. With the outstanding performance of LLMs, many studies have employed them for misinformation detection. Unfortunately, they focus on in-domain tasks and do not incorporate significant sentiment and emotion features (which we jointly call affect). In this paper, we propose RAEmoLLM, the first retrieval augmented (RAG) LLMs framework to address cross-domain misinformation detection using in-context learning based on affective information. RAEmoLLM includes three modules. (1) In the index construction module, we apply an emotional LLM to obtain affective embeddings from all domains to construct a retrieval database. (2) The retrieval module uses the database to recommend top K examples (text-label pairs) from source domain data for target domain contents. (3) These examples are adopted as few-shot demonstrations for the inference module to process the target domain content. The RAEmoLLM can effectively enhance the general performance of LLMs in cross-domain misinformation detection tasks through affect-based retrieval, without fine-tuning. We evaluate our framework on three misinformation benchmarks. Results show that RAEmoLLM achieves significant improvements compared to the other few-shot methods on three datasets, with the highest increases of 15.64%, 31.18%, and 15.73% respectively. This project is available at https://github.com/lzw108/RAEmoLLM.

Xiaodong Wu, Minhao Wang, Yichen Liu, Xiaoming Shi, He Yan, Lu Xiangju, Junmin Zhu, Wei Zhang

As Large Language Models (LLMs) evolve in natural language processing (NLP), their ability to stably follow instructions in long-context inputs has become critical for real-world applications. However, existing benchmarks seldom focus on instruction-following in long-context scenarios or stability on different inputs. To bridge this gap, we introduce LIFBench, a scalable dataset designed to evaluate LLMs’ instruction-following capabilities and stability across long contexts. LIFBench comprises three long-context scenarios and eleven diverse tasks, featuring 2,766 instructions generated through an automated expansion method across three dimensions: length, expression, and variables. For evaluation, we propose LIFEval, a rubric-based assessment method that enables precise, automated scoring of complex LLM responses without reliance on LLM-assisted assessments or human judgment. This method allows for a comprehensive analysis of model performance and stability from multiple perspectives. We conduct detailed experiments on 20 prominent LLMs across six length intervals. Our work contributes LIFBench and LIFEval as robust tools for assessing LLM performance in complex and long-context settings, offering valuable insights to guide future advancements in LLM development.

Dingyi Yang, Qin Jin

In this work, we conduct systematic research in a challenging area: the automatic evaluation of book-length stories (>100K tokens). Our study focuses on two key questions: (1) understanding which evaluation aspects matter most to readers, and (2) exploring effective methods for evaluating lengthy stories. We introduce the first large-scale benchmark, **LongStoryEval**, comprising 600 newly published books with an average length of 121K tokens (maximum 397K). Each book includes its average rating and multiple reader reviews, presented as critiques organized by evaluation aspects. By analyzing all user-mentioned aspects, we propose an *evaluation criteria structure* and conduct experiments to identify the most significant aspects among the 8 top-level criteria. For evaluation methods, we compare the effectiveness of three types: *aggregation-based*, *incremental-updated*, and *summary-based* evaluations. Our findings reveal that aggregation- and summary-based evaluations perform better, with the former excelling in detail assessment and the latter offering greater efficiency. Building on these insights, we further propose **NovelCritique**, an 8B model that leverages the efficient summary-based framework to review and score stories across specified aspects. NovelCritique outperforms commercial models like GPT-4o in aligning with human evaluations. All our datasets and codes will be released to foster further research.

Ercong Nie, Bo Shao, Mingyang Wang, Zifeng Ding, Helmut Schmid, Hinrich Schuetze

This paper introduces BMIKE-53, a comprehensive benchmark for cross-lingual in-context knowledge editing (IKE), spanning 53 languages and three KE datasets: zsRE, CounterFact, and WikiFactDiff. Cross-lingual KE, which requires knowledge edited in one language to generalize across diverse languages while preserving unrelated knowledge, remains underexplored. To address this, we systematically evaluate IKE under zero-shot, one-shot, and few-shot setups, including tailored metric-specific demonstrations. Our findings reveal that model scale and demonstration alignment critically govern cross-lingual editing efficacy, with larger models and tailored demonstrations significantly improving performance. Linguistic properties, particularly script type, strongly influence outcomes, with non-Latin languages underperforming due to issues like language confusion.

Zhiyuan Zhu, Yusheng Liao, Zhe Chen, Yuhao Wang, Yunfeng Guan, Yanfeng Wang, Yu Wang

Large language models (LLMs) are trained on extensive historical corpora, but their ability to understand time and maintain temporal awareness of time-evolving factual knowledge remains limited. Previous studies often neglect the critical aspect of utilizing knowledge from various sources. To address this gap, we introduce EvolveBench, a comprehensive benchmark that evaluates temporal competence along five key dimensions: Cognition, which examines the ability to recall and contextualize historical facts. Awareness, which tests LLMs’ awareness of temporal misalignment between external inputs and the temporal context of a query. Trustworthiness, which assesses whether models can identify and appropriately refuse queries based on invalid timestamps. Understanding, which focuses on interpreting both explicit dates and implicit historical markers. Finally, reasoning evaluates the capacity to analyze temporal relationships and draw accurate inferences. Evaluating 15 widely used LLMs on EvolveBench shows that GPT-4o achieves the highest average EM score of 79.36, while the open-source Llama3.1-70B demonstrates notable strength in handling temporally misaligned contexts with an average score of 72.47. Despite these advances, all models still struggle with handling temporal misaligned context. Our code and dataset are available at https://github.com/zzysjtuiwct/EvolveBench.

Romain Meunier, Farah Benamara, Véronique Moriceau, Zhongzheng Qiao, Savitha Ramasamy

This paper proposes CrisisTS, the first multimodal and multilingual dataset for urgency classification composed of benchmark crisis datasets from French and English social media about various expected (e.g., flood, storm) and sudden (e.g., earthquakes, explosions) crises that have been mapped with open source geocoded meteorological time series data. This mapping is based on a simple and effective strategy that allows for temporal and location alignment even in the absence of location mention in the text. A set of multimodal experiments have been conducted relying on transformers and LLMs to improve overall performances while ensuring model generalizability. Our results show that modality fusion outperforms text-only models.

Arduin Findeis, Floris Weers, Guoli Yin, Ke Ye, Ruoming Pang, Tom Gunter

Pairwise preferences over model responses are widely collected to evaluate and provide feedback to large language models (LLMs). Given two alternative model responses to the same input, a human or AI annotator selects the “better” response. This approach can provide feedback for domains where other hard-coded metrics are difficult to obtain (e.g., chat response quality), thereby helping model evaluation or training. However, for some domains high-quality pairwise comparisons can be tricky to obtain - from AI and humans. For example, for responses with many factual statements, annotators may disproportionately weigh writing quality rather than underlying facts. In this work, we explore augmenting standard AI annotator systems with additional tools to improve performance on three challenging response domains: long-form factual, math and code tasks. We propose a tool-using agentic system to provide higher quality feedback on these domains. Our system uses web-search and code execution to ground itself based on external validation, independent of the LLM’s internal knowledge and biases. We provide extensive experimental results evaluating our method across the three targeted response domains as well as general annotation tasks, using RewardBench (incl. AlpacaEval and LLMBar), RewardMath, as well as three new datasets for domains with saturated pre-existing datasets. Our results indicate that external tools can indeed improve performance in many, but not all, cases. More generally, our experiments highlight the sensitivity of performance to simple parameters (e.g., prompt) and the need for improved (non-saturated) annotator benchmarks. We share our code at https://github.com/apple/ml-agent-evaluator.

Kechi Zhang, Ge Li, Yihong Dong, Jingjing Xu, Jun Zhang, Jing Su, Yongfei Liu, Zhi Jin

Code generation models have shown significant potential for programming tasks. However, existing training methods like supervised fine-tuning face key limitations: they do not effectively teach models to prioritize correct over incorrect solutions in ambiguous situations, nor do they effectively optimize the runtime efficiency of the generated code. To address these challenges, we propose CodeDPO, a framework that integrates preference learning into code generation to improve two key code preference factors: code correctness and efficiency. CodeDPO employs a novel dataset construction method, utilizing a self-generation-and-validation mechanism that simultaneously generates and evaluates code and test cases. The underlying assumption is that test cases executable by multiple code snippets provide more reliable validation, and code that passes more tests is more likely to be correct. Through this self-validation process, our PageRank-inspired algorithm iteratively updates the ranking score of each code snippet, ultimately creating a code preference optimization dataset based on correctness and efficiency. CodeDPO is flexible and scalable, generating diverse preference optimization data without depending on powerful models such as GPT-4. Through comprehensive evaluations of five widely used benchmarks, CodeDPO demonstrates significant improvements in correctness and efficiency compared to existing methods. Our experiments prove that CodeDPO enhances the capabilities of LLMs in code generation and provides a robust foundation for conducting code preference optimization in more complex and challenging real-world scenarios.

Chao Wen, Jacqueline Staub, Adish Singla

Large language and multimodal models have shown remarkable success on various benchmarks focused on specific skills such as general-purpose programming, math word problem-solving, and visual question answering. However, it is unclear how well these models perform on tasks that require a combination of these skills. In this paper, we curate a novel program synthesis benchmark based on the real-world tasks in the XLogoOnline visual programming environment. Each task requires a combination of different skills such as spatial planning, basic programming, and logical reasoning. Our evaluation shows that current state-of-the-art models like GPT-4V and Llama3-70B struggle to solve these tasks, achieving only 20% and 2.35% success rates, respectively. Next, we develop a fine-tuning pipeline to boost the performance of models by leveraging a large-scale synthetic training dataset with over 80,000 tasks. Moreover, we showcase how emulator-driven feedback can be used to design a curriculum over training data distribution, through which a fine-tuned Llama3-8B drastically outperforms GPT-4V and Llama3-70B models. Finally, we provide an in-depth failure analysis to understand the limitations of different models. We will publicly release the benchmark for future research on program synthesis in visual programming.

Zichen Tang, Haihong E, Ziyan Ma, Haoyang He, Jiacheng Liu, Zhongjun Yang, Zihua Rong, Rongjin Li, Kun Ji, Qing Huang 等

We introduce **FinanceReasoning**, a novel benchmark designed to evaluate the reasoning capabilities of large reasoning models (LRMs) in financial numerical reasoning problems. Compared to existing benchmarks, our work provides three key advancements. (1) **Credibility**: We update 15.6% of the questions from four public datasets, annotating 908 new questions with detailed Python solutions and rigorously refining evaluation standards. This enables an accurate assessment of the reasoning improvements of LRMs. (2) **Comprehensiveness**: FinanceReasoning covers 67.8% of financial concepts and formulas, significantly surpassing existing datasets. Additionally, we construct 3,133 Python-formatted functions, which enhances LRMs’ financial reasoning capabilities through refined knowledge (*e.g.*, 83.2% \rightarrow 91.6% for GPT-4o). (3) **Challenge**: Models are required to apply multiple financial formulas for precise numerical reasoning on 238 *Hard* problems. The best-performing model (*i.e.*, OpenAI o1 with PoT) achieves 89.1% accuracy, yet LRMs still face challenges in numerical precision. We demonstrate that combining Reasoner and Programmer models can effectively enhance LRMs’ performance (*e.g.*, 83.2% \rightarrow 87.8% for DeepSeek-R1). Our work paves the way for future research on evaluating and improving LRMs in domain-specific complex reasoning tasks.

Irina Nikishina, Saba Anwar, Nikolay Dolgov, Maria Manina, Daria Ignatenko, Artem Shelmanov, Chris Biemann

Comparative Question Answering (CQA) lies at the intersection of Question Answering, Argument Mining, and Summarization. It poses unique challenges due to the inherently subjective nature of many questions and the need to integrate diverse perspectives. Although the CQA task can be addressed using recently emerged instruction-following Large Language Models (LLMs), challenges such as hallucinations in their outputs and the lack of transparent argument provenance remain significant limitations.To address these challenges, we construct a manually curated dataset comprising arguments annotated with their relevance. These arguments are further used to answer comparative questions, enabling precise traceability and faithfulness. Furthermore, we define explicit criteria for an “ideal” comparison and introduce a benchmark for evaluating the outputs of various Retrieval-Augmented Generation (RAG) models with respect to argument relevance. All code and data are publicly released to support further research.

Jiaheng Liu, Ken Deng, Congnan Liu, Jian Yang, Shukai Liu, He Zhu, Peng Zhao, Linzheng Chai, Yanan Wu, JinKe JinKe 等

Repository-level code completion has drawn great attention in software engineering, and several benchmarks have been introduced. However, existing repository-level code completion benchmarks usually focus on a limited number of languages (<5), which cannot evaluate the general code intelligence abilities across different languages for existing code Large Language Models (LLMs). Besides, the existing benchmarks usually report overall average scores of different languages, where the fine-grained abilities in different completion scenarios are ignored. Therefore, to facilitate the research of code LLMs in multilingual scenarios, we propose a massively multilingual repository-level code completion benchmark covering 18 programming languages (called M2RC-EVAL), and two types of fine-grained annotations (i.e., bucket-level and semantic-level) on different completion scenarios are provided, where we obtain these annotations based on the parsed abstract syntax tree. Moreover, we also curate a massively multilingual instruction corpora M2RC-INSTRUCT dataset to improve the repository-level code completion abilities of existing code LLMs. Comprehensive experimental results demonstrate the effectiveness of our M2RC-EVAL and M2RC-INSTRUCT.

Shuai Wang, Yinan Yu

While Large Language Models (LLMs) excel at many natural language processing tasks, they often suffer from factual inaccuracies in knowledge-intensive scenarios. Integrating external knowledge resources, particularly knowledge graphs (KGs), provides a transparent and updatable foundation for more reliable reasoning. Knowledge Base Question Answering (KBQA), which queries and reasons over KGs, is central to this effort, especially for complex, multi-hop queries. However, multi-hop reasoning poses two key challenges: (1) maintaining coherent reasoning paths, and (2) avoiding prematurely discarding critical multi-hop connections. To address these issues, we introduce iQUEST, a question-guided KBQA framework that iteratively decomposes complex queries into simpler sub-questions, ensuring a structured and focused reasoning trajectory. Additionally, we integrate a Graph Neural Network (GNN) to look ahead and incorporate 2-hop neighbor information at each reasoning step. This dual approach strengthens the reasoning process, enabling the model to explore viable paths more effectively. Detailed experiments demonstrate the consistent improvement delivered by iQUEST across four benchmark datasets and four LLMs.

Haozhe Xu, Xiaohua Wang, Changze Lv, Xiaoqing Zheng

Conversational recommender systems (CRSs) enhance recommendation quality by engaging users in multi-turn dialogues, capturing nuanced preferences through natural language interactions. However, these systems often face the false negative issue, where items that a user might like are incorrectly labeled as negative during training, leading to suboptimal recommendations. Expanding the label set through data augmentation presents an intuitive solution but faces the challenge of balancing two key aspects: ensuring semantic relevance and preserving the collaborative information inherent in CRS datasets. To address these issues, we propose a novel data augmentation framework that first leverages an LLM-based semantic retriever to identify diverse and semantically relevant items, which are then filtered by a relevance scorer to remove noisy candidates. Building on this, we introduce a two-stage training strategy balancing semantic relevance and collaborative information. Extensive experiments on two benchmark datasets and user simulators demonstrate significant and consistent performance improvements across various recommenders, highlighting the effectiveness of our approach in advancing CRS performance.

HangChen HangChen, Chao-Han Huck Yang, Jia-Chen Gu, Sabato Marco Siniscalchi, Jun Du

We introduce MISP-Meeting, a new real-world, multimodal dataset that covers subject-oriented long-form content. MISP-Meeting integrates information from speech, vision, and text modalities to facilitate automatic meeting transcription and summarization (AMTS). Challenging conditions in human meetings, including far-field speech recognition, audio-visual understanding, and long-term summarization, have been carefully evaluated. We benchmark state-of-the-art automatic speech recognition (ASR) and large language models (LLMs) on this dataset, enhanced with multimodal cues. Experiments demonstrate that incorporating multimodal cues, such as lip movements and visual focus of attention, significantly enhances transcription accuracy, reducing the character error rate (CER) from 36.60% to 20.27% via guided source separation (GSS), fine-tuning, and audio-visual fusion. Furthermore, our summarization analysis reveals a direct correlation between ASR quality and summary coherence, underscoring the importance of robust multimodal modeling. Our dataset and codebase will be released as open source.

Wanli Yang, Fei Sun, Jiajun Tan, Xinyu Ma, Qi Cao, Dawei Yin, Huawei Shen, Xueqi Cheng

Despite near-perfect results reported in the literature, the effectiveness of model editing in real-world applications remains unclear. To bridge this gap, we introduce QAEdit, a new benchmark aligned with widely used question answering (QA) datasets, and WILD, a task-agnostic evaluation framework designed to better reflect real-world usage of model editing. Our single editing experiments show that current editing methods perform substantially worse than previously reported (38.5% vs. 96.8%). We demonstrate that it stems from issues in the synthetic evaluation practices of prior work. Among them, the most severe is the use of teacher forcing during testing, which leaks both content and length of the ground truth, leading to overestimated performance. Furthermore, we simulate practical deployment by sequential editing, revealing that current approaches fail drastically with only 1000 edits. This work calls for a shift in model editing research toward rigorous evaluation and the development of robust, scalable methods that can reliably update knowledge in LLMs for real-world use.

Xueru Wen, Jie Lou, Zichao Li, Yaojie Lu, Xing Yu, Yuqiu Ji, Guohai Xu, Hongyu Lin, Ben He, Xianpei Han 等

Reward models (RMs) are crucial for aligning large language models (LLMs) with human preferences. However, most RM research is centered on English and relies heavily on synthetic resources, which leads to limited and less reliable datasets and benchmarks for Chinese. To address this gap, we introduce CheemsBench, a fully human-annotated RM evaluation benchmark within Chinese contexts, and CheemsPreference, a large-scale and diverse preference dataset annotated through human-machine collaboration to support Chinese RM training. We systematically evaluate open-source discriminative and generative RMs on CheemsBench and observe significant limitations in their ability to capture human preferences in Chinese scenarios. Additionally, based on CheemsPreference, we construct an RM that achieves state-of-the-art performance on CheemsBench, demonstrating the necessity of human supervision in RM training. Our findings reveal that scaled AI-generated data struggles to fully capture human preferences, emphasizing the importance of high-quality human supervision in RM development.

Kai He, Yucheng Huang, Wenqing Wang, Delong Ran, Dongming Sheng, Junxuan Huang, Qika Lin, Jiaxing Xu, Wenqiang Liu, Mengling Feng

This study introduces Crab, a novel Configurable Role-Playing (RP) LLM with Assessing Benchmark, which consists of Role-Centric Dataset Curation, Persona-Embodying LLM Construction, and Comprehensive Benchmark Creation for RP dialogue generation. Distinct from traditional RP models that employ only several preset roles, Crab enables dynamic configuration of desired roles, thereby enhancing related flexibility and adaptability. To effectively train RP-LLMs, we curated the largest RP training dataset. The dataset provides a detailed role overview for each dialogue, including character profile, conversation scenario, and tagged topic, capturing a broad range of role-based behaviors, emotions, and interactions. We also noticed that current benchmarks lack both proper evaluation standards and methods. Thus, to validate RP-LLMs’ effectiveness, we introduced a new benchmark containing an evaluation standard, a test dataset with manual annotations, and a reward model RoleRM designed to automatically assess specific aspects of RP while aligning with human perception. Sufficient experiments reveal that RoleRM significantly outperforms ChatGPT and other evaluation methods in conducting fine-grained evaluations of RP. Also, RP-LLMs powered by Crab demonstrate superior performance across various fine-grained aspects.

Tan Yue, Rui Mao, Xuzhao Shi, Shuo Zhan, Zuhao Yang, Dongyan Zhao

Question answering (QA) tasks serve as a key benchmark for evaluating generation systems. Traditional rule-based metrics, such as accuracy and relaxed-accuracy, struggle with open-ended and unstructured responses. LLM-based evaluation methods offer greater flexibility but suffer from sensitivity to instructions, robustness issues, and high computational costs. To overcome these challenges, we introduce QAEval, a hybrid framework combining rule-based reliability with LLM-based adaptability. QAEval utilizes two high-quality datasets: QAExtract for short-answer extraction and QAScore for scoring model training. By integrating a Mixture of Evaluators model with Dynamic Load Balancing Optimization, QAEval enables accurate, cost-effective QA evaluation. Experimental results show it outperforms models like GPT-4o and Claude-3, achieving 92.3% accuracy with only 0.6B parameters.

Lei Wang, Ruobing Zuo, Gaolei He, Jianlin Wang, Zhengfeng Yang

Automated Theorem Proving is an important and challenging task. Although large language models (LLMs) have demonstrated remarkable potential in mathematical reasoning, their performance in formal theorem proving remains constrained by the scarcity of high-quality supervised fine-tuning (SFT) data. To address this limitation, we propose a **Q**uality-**D**riven **T**heorem **S**ynthesis method (QDTSynth) in Lean4. During the statement synthesis, we enhance Monte Carlo Tree Search (MCTS) with an adaptive adjustment mechanism that dynamically optimizes the search strategy based on the synthesis of statements. In addition, we propose diversity screening and the self-assessment method to select theorems that exhibit both diversity and high quality from the initially synthetic statements, enabling the synthesis of a high-quality Lean4 theorem dataset. After fine-tuning three open-source large language models on our synthetic dataset, experiments on the miniF2F benchmark demonstrate that QDTSynth significantly improves the performance of various open-source LLMs in theorem proving tasks. Our work offers a promising new direction for the future synthesis of high-quality formal mathematical theorems.