论文检索

Hao Fang, Yuejie Zhang, Rui Feng, Yingwen Wang, Qing Wang, Wen He, Xiaobo Zhang, Tao Zhang, Shang Gao

Document-level relation extraction (DocRE) identifies relations between entities across an entire document. However, as the number and complexity of entities and entity-pair relations grow, the problem space expands quadratically, causing incomplete annotations and frequent false negatives, especially in biomedical datasets due to high construction costs. This leads to low recall in real-world scenarios. To address this, we propose GLiM, a novel framework that reduces the problem space using a graph-enhanced Transformer-based model and leverages large language models (LLMs) for reasoning. GLiM employs a cascaded approach: first, a graph-enhanced Transformer processes entity-pair relations with finer granularity by dynamically adjusting the graph size based on the number of entities; then, LLM inference handles challenging cases. Experiments show that GLiM boosts average recall and F1 scores by +6.34 and +4.41, respectively, outperforming state-of-the-art models on biomedical benchmarks. These results demonstrate the effectiveness of combining graph-enhanced Transformers with LLM inference for biomedical DocRE. Code will be released at https://github.com/HaoFang10/GLiM.

Yiyun Zhou, Chang Yao, Jingyuan Chen

The scaling law of Large Language Models (LLMs) reveals a power-law relationship, showing diminishing return on performance as model scale increases. While training LLMs from scratch is resource-intensive, fine-tuning a pre-trained model for specific tasks has become a practical alternative. Full fine-tuning (FFT) achieves strong performance; however, it is computationally expensive and inefficient. Parameter-efficient fine-tuning (PEFT) methods, like LoRA, have been proposed to address these challenges by freezing the pre-trained model and adding lightweight task-specific modules. LoRA, in particular, has proven effective, but its application to multi-task scenarios is limited by interference between tasks. Recent approaches, such as Mixture-of-Experts (MOE) and asymmetric LoRA, have aimed to mitigate these issues but still struggle with sample scarcity and noise interference due to their fixed structure. In response, we propose CoLA, a more flexible LoRA architecture with an efficient initialization scheme, which introduces three collaborative strategies to enhance performance by better utilizing the quantitative relationships between matrices A and B. Our experiments demonstrate the effectiveness and robustness of CoLA, outperforming existing PEFT methods, especially in low-sample scenarios. Our data and code are fully publicly available: https://github.com/zyy-2001/CoLA.

Tianyi Ma, Yiyue Qian, Zehong Wang, Zheyuan Zhang, Chuxu Zhang, Yanfang Ye

As the market for illicit drugs remains extremely profitable, major online platforms have become direct-to-consumer intermediaries for illicit drug trafficking participants. These online activities raise significant social concerns that require immediate actions. Existing approaches to combat this challenge are generally impractical due to the scarcity of labeled samples and imbalance of classes in real-world applications. To this end, we propose a novel \textbf{L}arge \textbf{L}anguage Model-empowered \textbf{Het}erogeneous \textbf{G}raph Prompt Learning framework for illicit \textbf{D}rug \textbf{T}rafficking detection, called \textbf{LLM-HetGDT} that leverages LLM to facilitate heterogeneous graph neural networks (HGNNs) to effectively identify minority classes, i.e., drug trafficking participants, in the class-imbalanced scenarios. Specifically, we first pre-train HGNN over a contrastive pretext task to capture the inherent node and structure information over an unlabeled drug trafficking heterogeneous graph (HG). Afterward, to alleviate the class-imbalanced issue, we leverage LLMs to augment the HG by generating high-quality synthetic user nodes in the minority classes. Then, we fine-tune the soft prompts on the augmented HG to capture the important information in the minority classes for the downstream drug trafficking detection task. To comprehensively study online illicit drug trafficking activities, we collect a new HG dataset over Twitter, called Twitter-HetDrug. Extensive experiments on this dataset demonstrate the effectiveness, efficiency, and applicability of our proposed method by comparing it with state-of-the-art baseline methods. Our source code is available at https://github.com/GraphResearcher/LLM-HetGDT.

Kanzhi Cheng, Wenpo Song, Jiaxin Fan, Zheng Ma, Qiushi Sun, Fangzhi Xu, Chenyang Yan, Nuo Chen, Jianbing Zhang, Jiajun Chen

Image captioning has been a longstanding challenge in vision-language research. With the rise of LLMs, modern Vision-Language Models (VLMs) generate detailed and comprehensive image descriptions. However, benchmarking the quality of such captions remains unresolved. This paper addresses two key questions: (1) How well do VLMs actually perform on image captioning, particularly compared to humans? We built CapArena, a platform with over 6000 pairwise caption battles and high-quality human preference votes. Our Arena-style evaluation marks a milestone, showing that leading models like GPT-4o achieve or even surpass human performance, while most open-source models lag behind. (2) Can automated metrics reliably assess caption quality? Using human annotations from CapArena, we evaluate traditional and recent captioning metrics, as well as VLM-as-a-Judge. Our analysis reveals that while some metrics (e.g., METEOR) show high caption-level agreement with humans, their systematic biases lead to inconsistencies in model ranking. In contrast, VLM-as-a-Judge demonstrates robust discernment at both the caption and model levels. Building on these insights, we release CapArena-Auto, an accurate and efficient automated benchmark for detailed captioning, achieving 93.4% correlation with human rankings at just $4 per test. All data and evaluation resources have been open-sourced.

Yuxi Sun, Aoqi Zuo, Wei Gao, Jing Ma

Large Language Models (LLMs) often exhibit knowledge disparities across languages. Encouraging LLMs to \textit{abstain} when faced with knowledge gaps is a promising strategy to reduce hallucinations in multilingual settings. Current abstention strategies for multilingual scenarios primarily rely on generating feedback in various languages using LLMs and performing self-reflection. However, these methods can be adversely impacted by inaccuracies and biases in the generated feedback. To address this, from a causal perspective, we introduce \textit{CausalAbstain}, a method that helps LLMs determine whether to utilize multiple generated feedback responses and how to identify the most useful ones. Extensive experiments demonstrate that \textit{CausalAbstain} effectively selects helpful feedback and enhances abstention decisions with interpretability in both native language (\textit{Casual-native}) and multilingual (\textit{Causal-multi}) settings, outperforming strong baselines on two benchmark datasets covering encyclopedic and commonsense knowledge QA tasks.

Zhaoxi Zhang, Sanwoo Lee, Zhixiang Wang, Yunfang Wu

The rapid advancement of Multi-modal Language Models (MLLMs) has significantly enhanced performance in multimodal tasks, yet these models often exhibit inherent biases that compromise their reliability and fairness. Traditional debiasing methods face a trade-off between the need for extensive labeled datasets and high computational costs. Model merging, which efficiently combines multiple models into a single one, offers a promising alternative but its usage is limited to MLLMs with the same architecture. We propose 3DM, a novel framework integrating Distill, Dynamic Drop, and Merge to address these challenges. 3DM employs knowledge distillation to harmonize models with divergent architectures and introduces a dynamic dropping strategy that assigns parameter-specific drop rates based on their contributions to bias and overall performance. This approach preserves critical weights while mitigating biases, as validated on the MMSD2.0 sarcasm detection dataset. Our key contributions include architecture-agnostic merging, dynamic dropping, and the introduction of the Bias Ratio (BR) metric for systematic bias assessment. Empirical results demonstrate that 3DM outperforms existing methods in balancing debiasing and enhancing the overall performance, offering a practical and scalable solution for deploying fair and efficient MLLMs in real-world applications.

Chuxue Cao, Han Zhu, Jiaming Ji, Qichao Sun, Zhenghao Zhu, Wu Yinyu, Josef Dai, Yaodong Yang, Sirui Han, Yike Guo

With the growing prevalence of large language models (LLMs), the safety of LLMs has raised significant concerns. However, there is still a lack of definitive standards for evaluating their safety due to the subjective nature of current safety benchmarks. To address this gap, we conducted the first exploration of LLMs’ safety evaluation from a legal perspective by proposing the SafeLawBench benchmark. SafeLawBench categorizes safety risks into three levels based on legal standards, providing a systematic and comprehensive framework for evaluation. It comprises 24,860 multi-choice questions and 1,106 open-domain question-answering (QA) tasks. Our evaluation included 2 closed-source LLMs and 18 open-source LLMs using zero-shot and few-shot prompting, highlighting the safety features of each model. We also evaluated the LLMs’ safety-related reasoning stability and refusal behavior. Additionally, we found that a majority voting mechanism can enhance model performance. Notably, even leading SOTA models like Claude-3.5-Sonnet and GPT-4o have not exceeded 80.5% accuracy in multi-choice tasks on SafeLawBench, while the average accuracy of 20 LLMs remains at 68.8%. We urge the community to prioritize research on the safety of LLMs.

Xinyu Pang, Ruixin Hong, Hongming Zhang, Changshui Zhang

Web tasks, which involve processing data from online resources, challenge agents to generalize beyond fixed knowledge to unseen task contexts. Learning from experience, the ability to derive reusable patterns from past tasks, is crucial for improving generalization. However, existing methods focus on summarizing workflows, i.e., common sub-routines, which may introduce excessive low-level details that distract models. Additionally, the absence of task-specific objectives can lead to inconsistencies between workflows and future task queries, hindering reasoning performance. This paper seeks to mitigate these issues by proposing A^2, a framework that derives task-adaptive hierarchical abstraction to enhance web task reasoning. Our approach first extracts general-purpose semantic abstraction from past task-solution pairs. Combined with the next task query, this abstraction forms a task-adaptive episodic abstraction that guides subsequent reasoning. Experiments show that A^2 achieves superior performance with competitive cost-efficiency, improving success rates by 0.7% on Mind2web and 4.6% on Webarena.

Lishuai Hou, Zixiong Wang, Gaoyang Liu, Chen Wang, Wei Liu, Kai Peng

Machine Unlearning (MU) has emerged as a promising solution for removing the influence of data that an owner wishes to unlearn from Large Language Models (LLMs). However, existing MU methods, which require tuning the entire model parameters on the unlearned data with random labels or perturbed gradients, significantly degrade model utility, especially given the difficulty of accessing the original training data. This presents a key challenge: how can we achieve MU using only the unlearned data while preserving model utility?In this paper, we propose NeuMuter, a simple but effective MU method that eliminates the influence of unlearned data from LLMs by modulating the outputs of merely 1% of the neurons in the feed-forward network (FFN) modules within the Transformer blocks, minimizing disruption to the model’s performance. We design a trainable masking scheme that decouples the memorization of different training data within the neurons of LLMs, allowing us to precisely identify and modify neurons associated with the unlearned data. Through comprehensive evaluations on two benchmarks across four different LLMs, we demonstrate that modifying the outputs of a few fraction of the total neurons can effectively achieve MU while preserving the model’s utility across downstream tasks.

Meng Lu, Yuzhang Xie, Zhenyu Bi, Shuxiang Cao, Xuan Wang

Large language models (LLMs) excel in generating unstructured text. However, they struggle with producing structured output while maintaining accuracy in zero-shot information extraction (IE), such as named entity recognition (NER) and relation extraction (RE). To address these challenges, we propose CROSSAGENTIE, a multi-agent framework that enhances zero-shot IE through multi-agent LLM collaboration. CROSSAGENTIE refines LLM predictions iteratively through two mechanisms: intra-group cross-type debate, which resolves entity-label conflicts through context-based evidence and confidence aggregation, and inter-group cross-task debate, where NER and RE mutually refine outputs via bidirectional feedback. Furthermore, we introduce template fine-tuning, distilling high-confidence multi-agent outputs into a single model, significantly reducing inference cost while preserving accuracy. Experiments across five NER and five RE datasets show that CROSSAGENTIE significantly outperforms state-of-the-art zero-shot baselines by a large margin. CROSSAGENTIE effectively addresses LLMs limitations in structured prediction with an effective and efficient approach for zero-shot information extraction.

Hongru Song, Yu-An Liu, Ruqing Zhang, Jiafeng Guo, Jianming Lv, Maarten de Rijke, Xueqi Cheng

We explore adversarial attacks against retrieval-augmented generation (RAG) systems to identify their vulnerabilities. We focus on generating human-imperceptible adversarial examples and introduce a novel imperceptible retrieve-to-generate attack against RAG. This task aims to find imperceptible perturbations that retrieve a target document, originally excluded from the initial top-k candidate set, in order to influence the final answer generation. To address this task, we propose ReGENT, a reinforcement learning-based framework that tracks interactions between the attacker and the target RAG and continuously refines attack strategies based on relevance-generation-naturalness rewards. Experiments on newly constructed factual and non-factual question-answering benchmarks demonstrate that ReGENT significantly outperforms existing attack methods in misleading RAG systems with small imperceptible text perturbations.

Guojun Xiong, Zhiyang Deng, Keyi Wang, Yupeng Cao, Haohang Li, Yangyang Yu, Xueqing Peng, Mingquan Lin, Kaleb E Smith, Xiao-Yang Liu 等

Large language models (LLMs) fine-tuned on multimodal financial data have demonstrated impressive reasoning capabilities in various financial tasks. However, they often struggle with multi-step, goal-oriented scenarios in interactive financial markets, such as trading, where complex agentic approaches are required to improve decision-making. To address this, we propose FLAG-Trader, a unified architecture integrating linguistic processing (via LLMs) with gradient-driven reinforcement learning (RL) policy optimization, in which a partially fine-tuned LLM acts as the policy network, leveraging pre-trained knowledge while adapting to the financial domain through parameter-efficient fine-tuning. Through policy gradient optimization driven by trading rewards, our framework not only enhances LLM performance in trading but also improves results on other financial-domain tasks. We present extensive empirical evidence to validate these enhancements.

Hoang Tran Vuong, Tue Le, Tu Vu, Tung Nguyen, Linh Ngo Van, Sang Dinh, Thien Huu Nguyen

Recent advances in neural topic models (NTMs) have improved topic quality but still face challenges: weak document-topic alignment, high inference costs due to large pretrained language models (PLMs), and limited modeling of hierarchical topic structures. To address these issues, we introduce HiCOT (Hierarchical Clustering and Contrastive Learning with Optimal Transport for Neural Topic Modeling), a novel framework that enhances topic coherence and efficiency. HiCOT integrates Optimal Transport to refine document-topic relationships using compact PLM-based embeddings, captures semantic structure of the documents. Additionally, it employs hierarchical clustering combine with contrastive learning to disentangle topic-word and topic-topic relationships, ensuring clearer structure and better coherence. Experimental results on multiple benchmark datasets demonstrate HiCOT’s superior effectiveness over existing NTMs in topic coherence, topic performance, representation quality, and computational efficiency.

Mengqiao Liu, Tevin Wang, Cassandra A. Cohen, Sarah Li, Chenyan Xiong

Which large language model (LLM) is better? Every evaluation tells a story, but what do users really think about current LLMs? This paper presents CLUE, an LLM-powered interviewer that conducts in-the-moment user experience interviews, right after users interact with LLMs, and automatically gathers insights about user opinions from massive interview logs. We conduct a study with thousands of users to understand user opinions on mainstream LLMs, recruiting users to first chat with a target LLM and then be interviewed by CLUE. Our experiments demonstrate that CLUE captures interesting user opinions, e.g., the bipolar views on the displayed reasoning process of DeepSeek-R1 and demands for information freshness and multi-modality. Our code and data are at https://github.com/cxcscmu/LLM-Interviewer.

Guo Zhenyuan, Yi Shi, Wenlong Meng, Chen Gong, Chengkun Wei, Wenzhi Chen

Model merging is a widespread technology in large language models (LLMs) that integrates multiple task-specific LLMs into a unified one, enabling the merged model to inherit the specialized capabilities of these LLMs. Most task-specific LLMs are sourced from open-source communities and have not undergone rigorous auditing, potentially imposing risks in model merging. This paper highlights an overlooked privacy risk: *an unsafe model could compromise the privacy of other LLMs involved in the model merging*. Specifically, we propose *PhiMM*, a privacy attack approach that trains a phishing model capable of stealing privacy using a crafted privacy phishing instruction dataset. Furthermore, we introduce a novel model cloaking method that mimics a specialized capability to conceal attack intent, luring users into merging the phishing model. Once victims merge the phishing model, the attacker can extract personally identifiable information (PII) or infer membership information (MI) by querying the merged model with the phishing instruction. Experimental results show that merging a phishing model increases the risk of privacy breaches. Compared to the results before merging, PII leakage increased by 3.9% and MI leakage increased by 17.4% on average. We release the code of *PhiMM* through an anonymous link.

Shi Yu, Zhiyuan Liu, Chenyan Xiong

Web crawl is a main source of large language models’ (LLMs) pretraining data, but the majority of crawled web pages are discarded in pretraining due to low data quality. This paper presents Craw4LLM, an efficient web crawling method that explores the web graph based on the preference of LLM pretraining. Specifically, it leverages the influence of a webpage in LLM pretraining as the priority score of the web crawler’s scheduler, replacing the standard graph-connectivity-based priority. Our experiments on a web graph containing 900 million webpages from a commercial search engine’s index demonstrate the efficiency of Craw4LLM in obtaining high-quality pretraining data. With just 21% URLs crawled, LLMs pretrained on Craw4LLM data reach the same downstream performances of previous crawls, significantly reducing the crawling waste and alleviating the burdens on websites. Our code is publicly available at https://github.com/cxcscmu/Craw4LLM.

Jacob Devasier, Akshith Reddy Putta, Rishabh Mediratta, Chengkai Li

We propose a novel paradigm for automatic fact-checking that leverages frame semantics to enhance the structured understanding of claims and guide the process of fact-checking them. To support this, we introduce a pilot dataset of real-world claims extracted from PolitiFact, specifically annotated for large-scale structured data. This dataset underpins two case studies: the first investigates voting-related claims using the Vote semantic frame, while the second explores various semantic frames based on data sources from the Organisation for Economic Co-operation and Development (OECD). Our findings demonstrate the effectiveness of frame semantics in improving evidence retrieval and explainability for fact-checking. Finally, we conducted a survey of frames evoked in fact-checked claims, identifying high-impact frames to guide future work in this direction.

Tuochao Chen, Nicholas Scott Batchelder, Alisa Liu, Noah A. Smith, Shyamnath Gollakota

We introduce LlamaPIE, the first real-time proactive assistant designed to enhance human conversations through discreet, concise guidance delivered via hearable devices. Unlike traditional language models that require explicit user invocation, this assistant operates in the background, anticipating user needs without interrupting conversations. We address several challenges, including determining when to respond, crafting concise responses that enhance conversations, leveraging knowledge of the user for context-aware assistance, and real-time, on-device processing. To achieve this, we construct a semi-synthetic dialogue dataset and propose a two-model pipeline: a small model that decides when to respond and a larger model that generates the response. We evaluate our approach on real-world datasets, demonstrating its effectiveness in providing helpful, unobtrusive assistance. User studies with our assistant, implemented on Apple Silicon M2 hardware, show a strong preference for the proactive assistant over both a baseline with no assistance and a reactive AI assistant, highlighting the potential of LlamaPIE to enhance live conversations.

Xianren Zhang, Xianfeng Tang, Hui Liu, Zongyu Wu, Qi He, Dongwon Lee, Suhang Wang

Recent studies show LLMs struggle with complex instructions involving multiple constraints (e.g., length, format, sentiment). Existing research enhances open-source LLMs using closed-source guidance (e.g., GPT-4), but this heavily relies on generated data quality. An alternative is leveraging LLMs’ self-correction to refine responses for better constraint adherence. However, this is limited by the feedback quality, as we found LLMs cannot generate reliable feedback or detect errors. Moreover, the self-correction effectiveness relies on few-shot examples illustrating response modifications. As constraints in complex instructions are diverse, manually crafting such examples for each constraint type can be labor-intensive and sub-optimal. To address these two challenges, we propose the Divide-Verify-Refine (DVR) framework with three steps: (1) Divide complex instructions into single constraints and prepare appropriate tools; (2) Verify responses using tools that provide rigorous check and textual guidance (e.g., Python scripts for format checks or pre-trained classifiers for content analysis); (3) Refine: To maximize refinement effectiveness, we propose dynamic few-shot prompting, where a refinement repository collects successful refinements, and these examples are selectively retrieved for future refinements. Recognizing the lack of complexity in existing datasets, we create a new dataset of complex instructions. DVR doubles Llama3.1-8B’s constraint adherence and triples Mistral-7B’s performance.

Gregory Price, Shaomei Wu

Dyslexia can affect writing, leading to unique patterns such as letter and homophone swapping. As a result, text produced by people with dyslexia often differs from the text typically used to train natural language processing (NLP) models, raising concerns about their effectiveness for dyslexic users. This paper examines the fairness of four commercial machine translation (MT) systems towards dyslexic text through a systematic audit using both synthetically generated dyslexic text and real writing from individuals with dyslexia. By programmatically introducing various dyslexic-style errors into the WMT dataset, we present insights on how dyslexic biases manifest in MT systems as the text becomes more dyslexic, especially with real-word errors. Our results shed light on the NLP biases affecting people with dyslexia – a population that often relies on NLP tools as assistive technologies, highlighting the need for more diverse data and user representation in the development of foundational NLP models.