EMNLP 2025 Papers — Page 10
Conference on Empirical Methods in Natural Language Processing · 1809 papers
LLM-Driven Implicit Target Augmentation and Fine-Grained Contextual Modeling for Zero-Shot and Few-Shot Stance Detection
Yanxu Ji, Hongfei Lin (Dalian University Of Technology)
ClassificationTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Propose a two-stage framework: first, use LLM to mine and annotate implicit targets, then employ a dynamic multi-layer context attention network to adaptively model text and targets for zero-shot and few-shot stance detection.
LLM-Guided Co-Training for Text Classification
Md Mezbaur Rahman (University of Illinois Chicago), Cornelia Caragea (University of Illinois Chicago)
ClassificationLarge Language ModelText
🎯 What it does: Proposed a semi-supervised text classification framework named LG-COTRAIN, which generates pseudo labels based on LLM and performs weighted co-training, utilizing two models to dynamically interact and update sample importance weights during training;
LLM-Guided Semantic Relational Reasoning for Multimodal Intent Recognition
Qianrui Zhou (Tsinghua University), Hanlei Zhang (Tsinghua University)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityChain-of-ThoughtAudio
🎯 What it does: Proposes a semantic relation reasoning framework (LGSRR) guided by large language models (LLMs), which automatically mines fine-grained semantics and constructs three logical relationships (importance, complementarity, and inconsistency), enabling efficient reasoning for multimodal intent recognition.
LLM-Independent Adaptive RAG: Let the Question Speak for Itself
Maria Marina (AIRI), Viktor Moskvoretskii (EPFL)
RetrievalComputational EfficiencyTextRetrieval-Augmented Generation
🎯 What it does: Propose a lightweight LLM-independent adaptive retrieval method based on external information and evaluate its effectiveness on multiple datasets.
LLM-OREF: An Open Relation Extraction Framework Based on Large Language Models
Hongyao Tu (Xiamen University), Jinsong Su (Xiamen University)
Knowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Built an open relation extraction framework based on large language models that can automatically discover and predict new relations in test instances.
LLMs are Better Than You Think: Label-Guided In-Context Learning for Named Entity Recognition
Fan Bai (Johns Hopkins University), Mark Dredze (Johns Hopkins University)
RecognitionTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: Propose a zero-shot context learning method called DEER, which leverages label statistics for example retrieval and error reflection to enhance the performance of large language models (LLMs) in named entity recognition (NER).
LLMs as World Models: Data-Driven and Human-Centered Pre-Event Simulation for Disaster Impact Assessment
Lingyao Li (University of South Florida), Min Deng (Texas Tech University)
TransformerLarge Language ModelWorld ModelImageTextMultimodalityTabularRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Utilizing large language models (LLMs) as world models to conduct pre-event simulations of community perception of seismic damage before earthquakes, generating Modified Mercalli Intensity (MMI) predictions at postal code and county levels;
LLMs Behind the Scenes: Enabling Narrative Scene Illustration
Melissa Roemmele (Midjourney), Max Kreminski (Midjourney)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelImageText
🎯 What it does: Developed a pipeline that leverages LLMs to generate scene descriptions and drive text-to-image models to automatically generate story scene illustrations, and created the SCENEILLUSTRATIONS dataset based on this.
LLMs cannot spot math errors, even when allowed to peek into the solution
Kv Aditya Srivatsa, Ekaterina Kochmar (Mohamed bin Zayed University of Artificial Intelligence)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Investigated the ability of LLMs to identify errors in students' mathematical problem-solving steps, and evaluated their performance when provided with reference answers and corrected student answers.
LLMs Don’t Know Their Own Decision Boundaries: The Unreliability of Self-Generated Counterfactual Explanations
Harry Mayne (University of Oxford), Adam Mahdi (University of Oxford)
Explainability and InterpretabilityData-Centric LearningLarge Language ModelPrompt EngineeringTabular
🎯 What it does: Investigated the effectiveness and minimality of self-generated counterfactual explanations (SCE) from large language models (LLMs), and assessed their reliability in practical reasoning.
LM-Searcher: Cross-domain Neural Architecture Search with LLMs via Unified Numerical Encoding
Yuxuan Hu (Chinese University of Hong Kong MMLab), Hongsheng Li (Chinese University of Hong Kong MMLab)
Neural Architecture SearchTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringImageBiomedical DataBenchmarkAudio
🎯 What it does: Proposes LM-Searcher, a general framework that leverages large language models for cross-domain neural architecture search.
LMR-BENCH: Evaluating LLM Agent’s Ability on Reproducing Language Modeling Research
Shuo Yan (University of Texas at Dallas), Xinya Du (University of Texas at Dallas)
AI Code AssistantTransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: Proposed the LMR-BENCH benchmark for systematically evaluating the ability of large language model (LLM) agents to reproduce code from natural language processing (NLP) research papers.
LoCt-Instruct: An Automatic Pipeline for Constructing Datasets of Logical Continuous Instructions
Hongyu Sun (Nara Institute of Science and Technology), Taro Watanabe (Nara Institute of Science and Technology)
Data SynthesisTransformerLarge Language ModelTextSequentialBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Built a fully automated pipeline, LoCt-Pipeline, to generate the MSQA dataset LoCt-Instruct containing logically coherent, multi-turn instruction chains.
Logical Reasoning with Outcome Reward Models for Test-Time Scaling
Ramya Keerthy Thatikonda (Monash University), Ehsan Shareghi (Monash University)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkChain-of-Thought
🎯 What it does: This paper investigates test-time scaling in logical reasoning tasks using Outcome Reward Models (ORM), achieving re-ranking of reasoning outputs by scoring the final results of reasoning paths;
LogiCoL: Logically-Informed Contrastive Learning for Set-based Dense Retrieval
Yanzhen Shen (Stanford University), Dan Roth (University of Pennsylvania)
RetrievalTransformerContrastive LearningText
🎯 What it does: Propose LOGICOL, a framework that integrates logical consistency constraints into contrastive learning to enhance the retrieval performance of dense retrieval models on queries containing logical connectives (e.g., AND, OR, NOT).
LogicTree: Structured Proof Exploration for Coherent and Rigorous Logical Reasoning with Large Language Models
Kang He (Purdue University), Kaushik Roy (Purdue University)
TransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Proposed the LogicTree framework, using modular LLMs and tree search to achieve structured, traceable logical proof exploration.
LogiDynamics: Unraveling the Dynamics of Inductive, Abductive and Deductive Logical Inferences in LLM Reasoning
Tianshi Zheng (Hong Kong University of Science and Technology), Simon See (NVIDIA)
Explainability and InterpretabilityLarge Language ModelImageTextChain-of-Thought
🎯 What it does: Systematic experiments on the logical reasoning dynamics of System 1 (direct reasoning) and System 2 (inductive/deductive reasoning) in large language models;
Logit Space Constrained Fine-Tuning for Mitigating Hallucinations in LLM-Based Recommender Systems
Jianfeng Deng, Lin Liu (University Of South Australia)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningText
🎯 What it does: Proposes the Logit Space Constraints Fine-Tuning (LCFT) framework to reduce hallucination phenomena in large language models when applied to recommendation systems;
Logits-Based Finetuning
Jingyao Li (Chinese University of Hong Kong), Jiaya Jia (Hong Kong University of Science and Technology)
Knowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningBenchmark
🎯 What it does: Proposed a logits-based fine-tuning framework that combines the teacher model's logits with true labels to construct richer training objectives, thereby enhancing the reasoning performance of small LLMs.
Logos as a Well-Tempered Pre-train for Sign Language Recognition
Ilya Ovodov (SberAI), Alexander Nagaev (SberAI)
RecognitionRepresentation LearningTransformerImageVideoBenchmark
🎯 What it does: This paper constructs the largest full-scale ISLR dataset Logos for Russian Sign Language and investigates the impact of cross-lingual transfer learning and visual similar gesture (VSSigns) annotations on model performance.
Long Chain-of-Thought Fine-tuning via Understanding-to-Reasoning Transition
Chenxin An (University of Hong Kong), Lingpeng Kong (University of Hong Kong)
TransformerSupervised Fine-TuningPrompt EngineeringTextChain-of-Thought
🎯 What it does: Proposes the Understanding-to-Reasoning Transition (URT) fine-tuning framework, which first enables the model to incorporate partial chain-of-thought (CoT) in the input and learn to generate the remaining reasoning, thereby achieving training for long-chain reasoning;
Long-Form Information Alignment Evaluation Beyond Atomic Facts
Danna Zheng (University of Edinburgh), Jeff Z. Pan (University of Edinburgh)
Large Language ModelTextSequentialBenchmark
🎯 What it does: Proposes a new attack method for information alignment evaluation called the MontageLie benchmark and designed the DOVESCORE framework that considers event order.
Look Again, Think Slowly: Enhancing Visual Reflection in Vision-Language Models
Pu Jian (Institute of Automation, Chinese Academy of Sciences), Jiajun Zhang (Institute of Automation, Chinese Academy of Sciences)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningMultimodality
🎯 What it does: Proposed the concept of visual reflection and constructed the Reflection-V model through a two-stage training strategy (cold start data construction + RL based on visual attention rewards).
Look Beyond Feeling: Unveiling Latent Needs from Implicit Expressions for Proactive Emotional Support
Xing Fu (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Construct an active emotional support dialogue framework named COCOON, and fine-tune the Llama3 model to achieve emotion need recognition and support dialogues
Lookahead Q-Cache: Achieving More Consistent KV Cache Eviction via Pseudo Query
Yixuan Wang (Harbin Institute of Technology), Wanxiang Che (Harbin Institute of Technology)
Computational EfficiencyTransformerText
🎯 What it does: Propose the Lookahead Q-Cache (LAQ) framework, which uses pre-generated low-quality pseudo queries to approximate real inference queries during KV cache clearance, significantly improving the accuracy of cache retention;
Looking Beyond Text: Reducing Language Bias in Large Vision-Language Models via Multimodal Dual-Attention and Soft-Image Guidance
Haozhe Zhao (University Of Illinois Urbana Champaign), Minjia Zhang (University Of Illinois Urbana Champaign)
Representation LearningTransformerLarge Language ModelVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose the LACING framework, which mitigates language bias in large vision-language models by employing a multi-modal dual attention mechanism (MDA) and soft image guidance (SIG), thereby enhancing visual information utilization and reducing hallucinations.
LoRACoE: Improving Large Language Model via Composition-based LoRA Expert
Guanyu Li (Fudan University), Xuanjing Huang (Fudan University)
Computational EfficiencyTransformerLarge Language ModelMixture of ExpertsTextBenchmark
🎯 What it does: This paper proposes a new LoRA-based Mixture of Experts (LoRACoE) model, which constructs experts by dynamically combining the rank levels of LoRA parameters;
LORAXBENCH: A Multitask, Multilingual Benchmark Suite for 20 Indonesian Languages
Alham Fikri Aji (MBZUAI), Trevor Cohn (University of Melbourne)
Large Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Propose LORAXBENCH, covering 20 Indonesian indigenous languages (including formal/informal variants for three languages) and six NLP tasks (reading comprehension, open-domain question answering, natural language inference, causal reasoning, machine translation, cultural question answering)
LoSiA: Efficient High-Rank Fine-Tuning via Subnet Localization and Optimization
Xujia Wang (Tsinghua University), Bin Xu (Tsinghua University)
Computational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose LoSiA, a parameter-efficient fine-tuning method that dynamically identifies and optimizes subnetworks during the fine-tuning process;
LVLMs are Bad at Overhearing Human Referential Communication
Zhengxiang Wang (Stony Brook University), Susan Brennan (Stony Brook University)
RecognitionVision Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Investigated the recognition capability of large vision-language models (LVLM) when exposed to spontaneous human referential communication, evaluating their performance as overhearers in a matching task using a dialogue corpus.
LyapLock: Bounded Knowledge Preservation in Sequential Large Language Model Editing
Peng Wang (Chinese Academy of Sciences), Songlin Hu (Chinese Academy of Sciences)
OptimizationTransformerLarge Language ModelText
🎯 What it does: Introduce the LyapLock framework in large model continuous editing, utilizing Lyapunov optimization to maintain long-term knowledge accuracy and model stability.
M-ABSA: A Multilingual Dataset for Aspect-Based Sentiment Analysis
ChengYan Wu, Barbara Plank (LMU Munich)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper constructs a parallel triplet (aspect, category, sentiment) ABSA dataset M-ABSA containing 21 languages and 7 domains, and provides baseline experiments.
M-BRe: Discovering Training Samples for Relation Extraction from Unlabeled Texts with Large Language Models
Zexuan Li (Nanjing University of Aeronautics and Astronautics), Piji Li (Nanjing University of Aeronautics and Astronautics)
ClassificationData-Centric LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Built an M-BRe framework based on large language models for efficiently generating relation extraction training samples from unannotated text
M-LongDoc: A Benchmark For Multimodal Super-Long Document Understanding And A Retrieval-Aware Tuning Framework
Yew Ken Chia (Singapore University Of Technology And Design), Lidong Bing (Singapore University Of Technology And Design)
RetrievalContrastive LearningMultimodalityBenchmarkFinance RelatedRetrieval-Augmented Generation
🎯 What it does: Proposed the M-LongDoc dataset and an automated evaluation framework for assessing multimodal long document understanding, along with a retrieval-aware fine-tuning method.
M-Wanda: Improving One-Shot Pruning for Multilingual LLMs
Rochelle Choenni (University of Amsterdam), Ivan Titov (University of Amsterdam)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Investigated the impact of sparsification on multilingual large language models (LLMs) and proposed an M-Wanda one-shot unstructured pruning method based on language-aware activation statistics and hierarchical sparse allocation, aiming to significantly preserve multilingual performance while drastically reducing model size.
M2Edit: Locate and Edit Multi-Granularity Knowledge in Multimodal Large Language Model
Yang Zhou (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences), Jun Zhao (Key Laboratory of Cognition and Decision Intelligence for Complex Systems, Institute of Automation, Chinese Academy of Sciences)
Explainability and InterpretabilityRepresentation LearningLarge Language ModelVision Language ModelMultimodality
🎯 What it does: This paper proposes the multi-modal knowledge editing framework MLE, which can locate and edit knowledge at different granularities (entities, relations, actions) in multi-modal large language models, and collaboratively update the components of multi-modal models;
M3Retrieve: Benchmarking Multimodal Retrieval for Medicine
Arkadeep Acharya (Indian Institute of Technology Patna), Dr Priti Singh
RetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Constructed a large-scale multi-modal medical retrieval benchmark, M3Retrieve, covering 16+ medical fields, 5 types of retrieval tasks, over 500K documents, and 100K queries, and conducted a systematic evaluation of existing retrieval models.
MA-DPR: Manifold-aware Distance Metrics for Dense Passage Retrieval
Yifan Liu (University of Toronto), Scott Sanner (University of Toronto)
RetrievalText
🎯 What it does: In Dense Passage Retrieval (DPR), a manifold-based distance metric MA-DPR is proposed. First, a K-nearest neighbor (KNN) graph is constructed to capture the sub-manifold nonlinear structure in the embedding space. Then, the shortest path on this graph is used as the query-passage distance for retrieval ranking.
MA-GTS: A Multi-Agent Framework for Solving Complex Graph Problems in Real-World Applications
Zike Yuan (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
OptimizationComputational EfficiencyLarge Language ModelAgentic AITextGraphBenchmarkChain-of-Thought
🎯 What it does: Proposed the MA-GTS multi-agent framework, which can automatically convert text descriptions from real-world scenarios into structured graph models and achieve efficient graph problem solving through multi-layer collaboration;
MAC-Tuning: LLM Multi-Compositional Problem Reasoning with Enhanced Knowledge Boundary Awareness
Junsheng Huang (Hong Kong University of Science and Technology), Yi R. Fung (Hong Kong University of Science and Technology)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: To address the hallucination problem of large language models in multi-problem scenarios, the MAC-Tuning method is proposed. It first identifies the model's knowledge boundaries and automatically annotates confidence levels, then performs two-step fine-tuning separately on answers and confidence scores, achieving decoupled learning of answer prediction and confidence estimation.
Machine-generated text detection prevents language model collapse
George Drayson (University College London), Vasileios Lampos (University College London)
GenerationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper studies the impact of different decoding strategies on model collapse during recursive training of large language models (LLMs) and proposes a weighted importance resampling method based on machine-generated text detectors to prevent model collapse.
MADAWSD: Multi-Agent Debate Framework for Adversarial Word Sense Disambiguation
Kaiyuan Zhang (Qilu University of Technology (Shandong Academy of Sciences)), Wenpeng Lu (Qilu University of Technology (Shandong Academy of Sciences))
Large Language ModelAgentic AITextBenchmarkChain-of-Thought
🎯 What it does: Proposed the MADAWSD framework based on multi-agent debate, utilizing LLM agents to perform word sense disambiguation in contexts containing adversarial information;
MAgICoRe: Multi-Agent, Iterative, Coarse-to-Fine Refinement for Reasoning
Justin Chen, Mohit Bansal (UNC Chapel Hill)
Explainability and InterpretabilityComputational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: Proposed an adaptive multi-agent coarse-to-fine hierarchical refinement framework called MAGICORE, which dynamically allocates computational resources and iteratively improves answers in LLM inference tasks based on problem difficulty.
Mahānāma: A Unique Testbed for Literary Entity Discovery and Linking
Sujoy Sarkar (Indian Institute of Technology Kharagpur), Pawan Goyal (Indian Institute of Technology Kharagpur)
RecognitionRetrievalTransformerTextBenchmark
🎯 What it does: Constructed and publicly released 'Mah¯ an¯ma a'—a large-scale Sanskrit entity discovery and linking dataset based on the Indian epic 'Mah¯ abh¯ arata,' covering 109K entity mentions, 5.5K unique entities, and associations with English knowledge bases;
MAIN: Mutual Alignment Is Necessary for instruction tuning
Fanyi Yang (Peking University), Qi Zhang (Microsoft Corporation)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed and implemented a Mutual Alignment framework named MAIN to generate high-quality instruction-response data and enhance instruction tuning through alignment reinforcement.
MAKAR: a Multi-Agent framework based Knowledge-Augmented Reasoning for Grounded Multimodal Named Entity Recognition
Xinkui Lin (Chinese Academy of Sciences), Hongbo Xu (Chinese Academy of Sciences)
RecognitionTransformerLarge Language ModelReinforcement LearningAgentic AIVision Language ModelMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper proposes the MAKAR multi-agent framework, which leverages internal and external knowledge enhancement to achieve multi-modal named entity recognition and localization.
Making VLMs More Robot-Friendly: Self-Critical Distillation of Low-Level Procedural Reasoning
Chan Young Park (University of Washington), Yejin Choi (Stanford University)
Knowledge DistillationRobotic IntelligenceVision Language ModelImageVideoChain-of-Thought
🎯 What it does: Propose SelfReVision—a self-improvement framework based on a cycle of self-critique, revision, and verification—to enhance the execution feasibility of low-capacity vision-language models in robotic program planning;
Mapping semantic networks to Dutch word embeddings as a diagnostic tool for cognitive decline
Maithe van Noort (University of Amsterdam), Jelke Bloem
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelTextGraphBiomedical Data
🎯 What it does: This paper constructs a semantic network based on word embeddings from Dutch BERTje, FastText, and XLM-RoBERTa, extracts network distance metrics such as path length, average diameter, and centroid diameter, and compares the performance of these metrics with traditional word count methods in predicting MMSE cognitive scores.
Mapping the Minds of LLMs: A Graph-Based Analysis of Reasoning LLMs
Zhen Xiong (University of Southern California), Yiwei Wang (University of California Merced)
Explainability and InterpretabilityLarge Language ModelGraphChain-of-Thought
🎯 What it does: Constructed a complete process that converts the long Chain-of-Thought outputs from large reasoning models into semantic clustering steps, forming a directed graph, and subsequently quantifies the reasoning structure through graph metrics while correlating it with model performance.
Mapping Toxic Comments Across Demographics: A Dataset from German Public Broadcasting
Jan Fillies (Freie Universität Berlin), Adrian Paschke (Freie Universität Berlin)
ClassificationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper constructs and releases the first multi-platform, age-stratified German toxic comment dataset, combining human annotation with LLM expansion to explore toxic language characteristics across different age groups and platforms.
Massive Supervised Fine-tuning Experiments Reveal How Data, Layer, and Training Factors Shape LLM Alignment Quality
Yuto Harada (NII LLMC), Yu Takagi (Nagoya Institute of Technology)
Data-Centric LearningLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Conduct systematic experiments on over 1,000 SFT models to investigate how training data, layers, and training methods affect the alignment quality of LLMs.
MathTutorBench: A Benchmark for Measuring Open-ended Pedagogical Capabilities of LLM Tutors
Jakub Macina (ETH Zurich), Mrinmaya Sachan (ETH Zurich)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: Proposes the MathTutorBench benchmark for rapidly evaluating the teaching capabilities of large language models in mathematics tutoring, encompassing three dimensions: professional knowledge, student comprehension, and teaching skills, along with seven corresponding tasks;
Matter-of-Fact: A Benchmark for Verifying the Feasibility of Literature-Supported Claims in Materials Science
Peter Jansen (University of Arizona), Ruoyao Wang (University of Arizona)
TransformerTextBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed and constructed the MATTER-OF-FACT benchmark to assess the feasibility of scientific hypotheses in materials science;
MAviS: A Multimodal Conversational Assistant For Avian Species
Yevheniia Kryklyvets (Mohamed bin Zayed University of Artificial Intelligence), Hisham Cholakkal (Mohamed bin Zayed University of Artificial Intelligence)
Domain AdaptationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmarkAudio
🎯 What it does: Constructed the MAviS multimodal bird dataset, trained and evaluated the MAviS-Chat dialogue model, and released the MAviS-Bench benchmark.
MAVL: A Multilingual Audio-Video Lyrics Dataset for Animated Song Translation
Woohyun Cho (Yonsei University), Youngjae Yu (Yonsei University)
TransformerLarge Language ModelVideoTextMultimodalityBenchmarkChain-of-ThoughtAudio
🎯 What it does: This paper proposes the multilingual audio-video lyrics dataset MAVL, and develops the SylAVL-CoT model based on this dataset, achieving multimodal lyric translation.
MaZO: Masked Zeroth-Order Optimization for Multi-Task Fine-Tuning of Large Language Models
Zhen Zhang (University of California, Santa Barbara), Zheng Zhang (University of California, Santa Barbara)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed the MaZO framework, which achieves efficient fine-tuning of large language models for multi-task learning by utilizing mask-based zeroth-order optimization (only forward inference).
MCIP: Protecting MCP Safety via Model Contextual Integrity Protocol
Huihao Jing (Hong Kong University of Science and Technology), Yangqiu Song (Hong Kong University of Science and Technology)
Safty and PrivacyLarge Language ModelSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: Designed and implemented a more secure MCP version called MCIP, constructed safety risk classification, benchmarks, and training data, and evaluated the security performance of LLMs in MCP scenarios.
Measuring and Mitigating Media Outlet Name Bias in Large Language Models
Seong-Jin Park (Catholic University of Korea), Kang-Min Kim (Catholic University of Korea)
OptimizationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper systematically evaluates political bias in media institution names within large language models (LLMs) and proposes a unified quantitative metric, SIPS, to measure the magnitude and direction of bias. Subsequently, an automated prompt optimization framework is constructed using this metric, significantly reducing name bias in news bias prediction and summarization tasks.
Measuring Bias or Measuring the Task: Understanding the Brittle Nature of LLM Gender Biases
Bufan Gao (University of Chicago), Elisa Kreiss (University of California, Los Angeles)
Explainability and InterpretabilityLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper systematically modifies prompts (including whether instructions and gender salience prompts are included) to test six open-source LLMs on four bias evaluation tasks (completion, association, multiple-choice, and sentence cloze), investigating the impact of prompts on gender bias measurement results and quantifying model sensitivity under different prompt conditions.
Measuring Chain of Thought Faithfulness by Unlearning Reasoning Steps
Martin Tutek (Technion Israel Institute of Technology), Yonatan Belinkov (Technion Israel Institute of Technology)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper proposes the Parametric Faithfulness Framework (PFF), which quantitatively evaluates the faithfulness of a model's internal reasoning by applying machine unlearning (NPO+KL) to chain-of-thought steps.
Measuring Risk of Bias in Biomedical Reports: The RoBBR Benchmark
Jianyou Wang (University of California San Diego), Leon Bergen (University of California San Diego)
ClassificationRetrievalTransformerLarge Language ModelBiomedical DataBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Created the RoBBR benchmark to evaluate model performance in the task of assessing risk of bias in biomedical research, and designed a main task and two subtasks (Support Sentence Retrieval (SSR) and Support Judgment Selection (SJS)).
Measuring scalar constructs in social science with LLMs
Hauke Licht (University of Innsbruck), Alexander Miserlis Hoyle (ETH Zürich)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringContrastive LearningText
🎯 What it does: This paper investigates how to leverage large language models to quantify continuous constructs in social science texts, comparing the effectiveness of various prompting and fine-tuning methods.
Measuring the Effect of Disfluency in Multilingual Knowledge Probing Benchmarks
Kirill Semenov (University of Zurich), Rico Sennrich (University of Zurich)
RetrievalTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Investigate the impact of template filling and sentence-level translation on knowledge extraction from large models in multilingual fact retrieval benchmarks, with a particular focus on grammatical and lexical accuracy.
MEBench: Benchmarking Large Language Models for Cross-Document Multi-Entity Question Answering
Teng Lin (Hong Kong University of Science and Technology), Nan Tang (Hong Kong University of Science and Technology)
TransformerLarge Language ModelTextTabularBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose MEBench, a cross-document multi-entity question answering benchmark, to evaluate the capabilities of LLMs and RAG systems in information retrieval, merging, and reasoning.
Mechanisms vs. Outcomes: Probing for Syntax Fails to Explain Performance on Targeted Syntactic Evaluations
Ananth Agarwal (Stanford University), Shikhar Murty (Stanford University)
Explainability and InterpretabilityTransformerLarge Language ModelTextBenchmark
🎯 What it does: Systematically compare the syntax learning mechanisms of large language models with their performance in targeted syntax evaluation, exploring the explanatory power of linear probes on syntactic knowledge.
Med-PRM: Medical Reasoning Models with Stepwise, Guideline-verified Process Rewards
Jaehoon Yun (Korea University), Jaewoo Kang (Korea University)
Explainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerReinforcement LearningTextBiomedical DataBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose Med-PRM, a Retrieval-Augmented Process Reward Model, for step-by-step verification of medical reasoning steps.
Med-VRAgent: A Framework for Medical Visual Reasoning-Enhanced Agents
Guangfu Guo (University of Birmingham), Yue Feng (University of Birmingham)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AIVision Language ModelImageTextMultimodalityBiomedical DataRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Built a multimodal agent framework named Med‑VRAgent to improve visual reasoning and report generation for medical images.
MedFact: A Large-scale Chinese Dataset for Evidence-based Medical Fact-checking of LLM Responses
Tong Chen (Xi'an Jiaotong-Liverpool University), Jionglong Su (Xi'an Jiaotong-Liverpool University)
ClassificationData-Centric LearningTransformerLarge Language ModelTextBiomedical DataBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the MEDFACT dataset—the first evidence-driven fact-checking dataset for LLM-generated medical content, containing 1,321 medical questions and 7,409 verifiable statements.
MedHallu: A Comprehensive Benchmark for Detecting Medical Hallucinations in Large Language Models
Shrey Pandit (University of Texas at Austin), Ying Ding (University of Texas at Austin)
ClassificationAnomaly DetectionTransformerLarge Language ModelTextBiomedical DataBenchmark
🎯 What it does: Constructed the MedHallu medical hallucination detection benchmark, containing 10,000 question-answer pairs derived from PubMedQA, and generated high-quality hallucinated answers through multi-model voting and NLI filtering.
Media Source Matters More Than Content: Unveiling Political Bias in LLM-Generated Citations
Sunhao Dai (Renmin University of China), Tat-Seng Chua (National University of Singapore)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextRetrieval-Augmented Generation
🎯 What it does: This paper investigates whether large language models (LLMs) exhibit political bias when generating answers with in-text citations, and systematically evaluates the tendency of LLMs to cite left-leaning versus right-leaning media by constructing a new AllSides-2024 dataset.
MedLinkDE – MedDRA Entity Linking for German with Guided Chain of Thought Reasoning
Roman Christof (Goethe-University Frankfurt), Alexander Mehler (Goethe-University Frankfurt)
Data SynthesisRetrievalRepresentation LearningTransformerSupervised Fine-TuningContrastive LearningTextBiomedical DataChain-of-Thought
🎯 What it does: Propose the MedLinkDE method, which implements a two-step workflow for German MedDRA entity linking: retrieval embedding + Guided CoT re-ranking based on coding guidelines.
Membership and Memorization in LLM Knowledge Distillation
Ziqi Zhang (Peking University), Hamed Haddadi (Brave Software)
Safty and PrivacyKnowledge DistillationTransformerLarge Language ModelText
🎯 What it does: This paper investigates the risks of membership information and memorization privacy leakage during the knowledge distillation (KD) process of large language models, systematically evaluating six mainstream KD techniques;
MemeArena: Automating Context-Aware Unbiased Evaluation of Harmfulness Understanding for Multimodal Large Language Models
Zixin Chen (Beijing University of Posts and Telecommunications), Jing Ma (Hong Kong Baptist University)
Agentic AIPrompt EngineeringVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose the MemeArena framework, which utilizes multiple agents under contextual simulation and multi-perspective fusion to conduct unbiased, context-aware evaluation of harmful content understanding in multimodal large language models.
MemeIntel: Explainable Detection of Propagandistic and Hateful Memes
Mohamed Bayan Kmainasi (Qatar Computing Research Institute), Firoj Alam
ClassificationExplainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextMultimodality
🎯 What it does: Constructed the MemeXplain dataset, providing labels and natural explanations for Arabic promotional memes and English hate memes; proposed a multi-stage optimization training scheme to simultaneously enhance the performance of vision-language models (VLMs) in detection and explanation generation.
MemeReaCon: Probing Contextual Meme Understanding in Large Vision-Language Models
Zhengyi Zhao (Chinese University of Hong Kong), Xian Wu (Tencent)
ClassificationGenerationTransformerVision Language ModelMultimodalityBenchmark
🎯 What it does: Propose the MemeReaCon benchmark to evaluate the ability of large vision-language models to understand memes in their original context, and assess models through four tasks.
MemInsight: Autonomous Memory Augmentation for LLM Agents
Rana Salama (George Washington University), Yassine Benajiba (AWS AI Labs)
RetrievalRecommendation SystemTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Propose MemInsight, an autonomous memory-enhancing framework that generates structured semantic enhancements for historical interactions of large language model agents through attribute mining and prioritization, thereby improving retrieval and context understanding.
Memorization \neq Understanding: Do Large Language Models Have the Ability of Scenario Cognition?
Boxiang Ma (Shanxi University), Xiaoli Li (Singapore University of Technology and Design)
Data SynthesisRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes a dual-perspective (output and internal representation) evaluation framework to assess LLMs' situational cognition ability, and constructs a scenario-based fictional facts dataset.
Memorization or Reasoning? Exploring the Idiom Understanding of LLMs
Jisu Kim (Hanyang University), Taeuk Kim (Hanyang University)
TransformerLarge Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Proposed the MIDAS multilingual idiom dataset and evaluated the understanding of large language models (LLMs) of idioms using multiple dimensions;
Memory OS of AI Agent
Jiazheng Kang (Beijing University of Posts and Telecommunications), Ting Bai (Beijing University of Posts and Telecommunications)
Large Language ModelAgentic AITextRetrieval-Augmented Generation
🎯 What it does: Propose MemoryOS, a memory operating system for AI agents, which implements long-term conversation memory management through four modules: hierarchical storage, update, retrieval, and generation.
Memory-QA: Answering Recall Questions Based on Multimodal Memories
Hongda Jiang (Meta Reality Labs), Xin Luna Dong (Meta Reality Labs)
RetrievalSupervised Fine-TuningVision Language ModelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Proposed the Memory-QA task, studying how to record, retrieve, and answer recall questions based on multimodal memory, and built the PENSIEVE system to achieve end-to-end solutions.
MentalGLM Series: Explainable Large Language Models for Mental Health Analysis on Chinese Social Media
Wei Zhai (Beijing University of Technology), Guanghui Fu (Sorbonne Université)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBiomedical DataChain-of-Thought
🎯 What it does: Construct a Chinese social media mental health explainable instruction dataset, C-IMHI, and perform two-stage fine-tuning on the GLM open-source LLM, resulting in the MentalGLM series, which achieves multi-task mental health analysis and explainable outputs.
MEPT: Mixture of Expert Prompt Tuning as a Manifold Mapper
Runjia Zeng (Rochester Institute of Technology), Dongfang Liu (Rochester Institute of Technology)
ClassificationTransformerSupervised Fine-TuningPrompt EngineeringMixture of ExpertsText
🎯 What it does: Proposes a Mixture of Expert Prompt Tuning (MEPT) method, which dynamically selects expert prompts via a sparse Mixture-of-Experts router to achieve adaptive mapping for different task data manifolds.
Merge then Realign: Simple and Effective Modality-Incremental Continual Learning for Multimodal LLMs
Dingkun Zhang (Harbin Institute of Technology), Xuan Wang (Harbin Institute of Technology)
TransformerLarge Language ModelSupervised Fine-TuningImageVideoMultimodalityPoint CloudAudio
🎯 What it does: In Modality-Incremental Continual Learning (MCL), a two-stage method named MERA (Merge then ReAlign) is proposed to address the two major issues of forgetting and mismatch.
Merger-as-a-Stealer: Stealing Targeted PII from Aligned LLMs with Model Merging
Lin Lu (Huazhong University of Science and Technology), Pan Zhou (Huazhong University of Science and Technology)
Safty and PrivacyAdversarial AttackTransformerSupervised Fine-TuningText
🎯 What it does: Propose a two-phase attack framework called Merger-as-a-Stealer, which exploits the model fusion process to steal target PII from aligned LLMs.
MERMAID: Multi-perspective Self-reflective Agents with Generative Augmentation for Emotion Recognition
Zhongyu Yang (Lanzhou University), Yingfang Yuan (Heriot-Watt University)
RecognitionData SynthesisTransformerLarge Language ModelDiffusion modelImageMultimodality
🎯 What it does: Propose a multi-agent framework MERMAID that enhances image emotion recognition using multi-perspective self-reflection and generative augmentation;
MessIRve: A Large-Scale Spanish Information Retrieval Dataset
Francisco Valentini, Juan Manuel Pérez (CONICET-Universidad de Buenos Aires)
RetrievalSupervised Fine-TuningTextBenchmark
🎯 What it does: This paper constructs and releases a large-scale Spanish information retrieval dataset called MessIRve, containing approximately 700,000 queries from the Google Autocomplete API and their corresponding Wikipedia relevant documents;
Meta-Rewarding Language Models: Self-Improving Alignment with LLM-as-a-Meta-Judge
Tianhao Wu (University of California, Berkeley), Sainbayar Sukhbaatar (Meta FAIR)
Meta LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringText
🎯 What it does: Propose the Meta-Rewarding method, enabling a single LLM to simultaneously act as an actor, judge, and meta-judge, achieving unsupervised self-improvement through the meta-judge's evaluation of its own judgments.
Meta-Semantics Augmented Few-Shot Relational Learning
Han Wu (University of Sydney), Jie Yin (University of Sydney)
Representation LearningMeta LearningPrompt EngineeringContrastive LearningGraphBenchmark
🎯 What it does: To address few-shot relation learning in knowledge graphs, the PromptMeta framework is proposed, which enhances few-shot reasoning performance by combining meta-learning with semantic prompts.
MetaFaith: Faithful Natural Language Uncertainty Expression in LLMs
Gabrielle Kaili-May Liu (Yale University), Arman Cohan (Yale University)
Explainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: Investigate the 'faithful calibration' of LLMs, systematically evaluate 19 models, 10 datasets, and multiple prompting strategies, finding that most models exhibit poor alignment between internal uncertainty and natural language uncertainty. Propose MetaFaith—a black-box calibration method based on metacognitive prompting—which significantly improves faithful calibration across multiple tasks and domains.
METok: Multi-Stage Event-based Token Compression for Efficient Long Video Understanding
Mengyue Wang (Technical University of Munich), Yunpu Ma (LMU Munich)
CompressionComputational EfficiencyTransformerVideo
🎯 What it does: Proposes a training-agnostic, multi-stage, event-aware visual token compression framework called METok, designed to reduce redundant visual tokens and lower computational and memory overhead during video LLM inference
Metric Calculating Benchmark: Code-Verifiable Complicate Instruction Following Benchmark for Large Language Models
Hyeonseok Moon (Korea University), Heuiseok Lim (Korea University)
Large Language ModelTextBenchmark
🎯 What it does: Created a verifiable, code-based benchmark called MCBench for evaluating complex instruction following, mathematical reasoning, and long-range consistency in large language models.
MiCRo: Mixture Modeling and Context-aware Routing for Personalized Preference Learning
Jingyan Shen (New York University), Han Zhao (University of Illinois Urbana-Champaign)
Recommendation SystemReinforcement Learning from Human FeedbackLarge Language ModelMixture of Experts
🎯 What it does: This paper proposes the MiCRo two-stage framework, which learns personalized preferences from large-scale binary preference data through hybrid modeling and context-aware routing.
MicroEdit: Neuron-level Knowledge Disentanglement and Localization in Lifelong Model Editing
Shiqi Wang (Jilin University), Yi Chang (Jilin University)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningLarge Language ModelAuto EncoderTextBenchmark
🎯 What it does: Propose the MicroEdit framework, which utilizes sparse autoencoders to decompose and locate knowledge within LLM internal representations, achieving knowledge editing at the fine-tuning level in lifelong learning scenarios.
Middo: Model-Informed Dynamic Data Optimization for Enhanced LLM Fine-Tuning via Closed-Loop Learning
Zinan Tang (OpenDataLab), Lijun Wu (OpenDataLab)
OptimizationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed and implemented Middo, a closed-loop dynamic data optimization framework based on model self-diagnosis for supervised fine-tuning of large language models.
MiLQ: Benchmarking IR Models for Bilingual Web Search with Mixed Language Queries
Jonghwi Kim (POSTECH), Gary Lee
RetrievalKnowledge DistillationTransformerTextBenchmark
🎯 What it does: Propose MiLQ, the first publicly available multilingual query benchmark, and evaluate the performance of multilingual IR models in cross-lingual retrieval on this benchmark.
Mind the Blind Spots: A Focus-Level Evaluation Framework for LLM Reviews
Hyungyu Shin (KAIST), Juho Kim (KAIST)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposed an attention evaluation framework for LLM-generated paper reviews and implemented an automated evaluation pipeline;
Mind the Gap: A Closer Look at Tokenization for Multiple-Choice Question Answering with LLMs
Mario Sanz-Guerrero (Johannes Gutenberg University Mainz), Katharina von der Wense (Johannes Gutenberg University Mainz)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper systematically investigates the impact of tokenizing the space after a colon along with the answer letter together or separately on the accuracy and calibration of large language models (LLMs) in multiple-choice question (MCQA) evaluations.
Mind the Gap: How BabyLMs Learn Filler-Gap Dependencies
Chi-Yun Chang (University of Michigan), Huteng Dai (University of Michigan)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper investigates whether a GPT-2 model trained exclusively on child language data can learn and generalize filler-gap dependencies, and evaluates its performance under syntactic constraints such as island structures.
Mind the Inclusivity Gap: Multilingual Gender-Neutral Translation Evaluation with mGeNTE
Beatrice Savoldi (Fondazione Bruno Kessler), Luisa Bentivogli (Fondazione Bruno Kessler)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Proposed and made public the multilingual gender-neutral translation benchmark MGENTE, and systematically evaluated the performance of various open-source instruction-following language models on the gender-neutral translation (GNT) task using it.
Mind the Value-Action Gap: Do LLMs Act in Alignment with Their Values?
Hua Shen (New York University), Tanu Mitra (University of Washington)
Explainability and InterpretabilityLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper proposes the VALUE-ACTION LEENS framework to evaluate the consistency between the claimed values and specific actions of large language models (LLMs) across different cultural and social contexts, and constructs the VIA dataset containing 14.8k value-oriented behavioral samples;