EMNLP 2025 Papers — Page 6
Conference on Empirical Methods in Natural Language Processing · 1809 papers
Efficient Context Selection for Long-Context QA: No Tuning, No Iteration, Just Adaptive‐k
Chihiro Taguchi (University of Notre Dame), Nikita Bhutani (Megagon Labs)
RetrievalComputational EfficiencyTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes an adaptive, single-channel, unsupervised retrieval strategy called Adaptivek, which determines how many paragraphs to retrieve by analyzing the maximum gap in the similarity distribution between queries and candidate documents, achieving efficient long-text question answering.
Efficient Model Development through Fine-tuning Transfer
Pin-Jie Lin (Virginia Tech), Tu Vu (Virginia Tech)
Computational EfficiencyKnowledge DistillationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a method to transfer fine-tuning updates (diff vector) from old model versions to new model versions, achieving significant performance improvements without additional training; meanwhile, it explores the feasibility of this method in scenarios such as multilingual model development and iterative upgrades.
Efficient Real-time Refinement of Language Model Text Generation
Joonho Ko (Korea Advanced Institute Of Science And Technology), Sung Ju Hwang (Korea Advanced Institute Of Science And Technology)
GenerationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningTextRetrieval-Augmented Generation
🎯 What it does: Proposed StreamingVR, which performs real-time sentence-level verification and correction of text generated by large language models to prevent error propagation
Efficient Unstructured Pruning of Mamba State-Space Models for Resource-Constrained Environments
Ibne Farabi Shihab (Iowa State University), Anuj Sharma (Iowa State University)
Computational EfficiencyTextTime SeriesSequential
🎯 What it does: Proposed an unstructured pruning framework for the Mamba state space model, capable of reducing the parameter count by up to 70% while maintaining most of the performance.
EGOILLUSION: Benchmarking Hallucinations in Egocentric Video Understanding
Ashish Seth (University of Maryland College Park), Dinesh Manocha (University of Maryland College Park)
TransformerLarge Language ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: Propose the EGOILLUSION benchmark for systematic evaluation of hallucinations in multimodal large language models on first-person videos.
EIFBENCH: Extremely Complex Instruction Following Benchmark for Large Language Models
Tao Zou (Tongyi Lab, Alibaba Group), Yongbin Li (Tongyi Lab, Alibaba Group)
TransformerLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: Proposed an extremely complex instruction-following benchmark called EIFBENCH, designed a multi-task multi-constraint evaluation framework, and introduced the Segment Policy Optimization (SegPO) algorithm to enhance large language models' (LLMs) multi-instruction execution capabilities.
Eliciting Implicit Acoustic Styles from Open-domain Instructions to Facilitate Fine-grained Controllable Generation of Speech
Jianxing Yu (Sun Yat-sen University), Jian Yin (Sun Yat-sen University)
GenerationTransformerLarge Language ModelReinforcement LearningDiffusion modelContrastive LearningTextMultimodalityRetrieval-Augmented GenerationAudio
🎯 What it does: This paper proposes a controllable speech generation method based on open-domain natural language instructions, which utilizes a multimodal large language model to infer implicit acoustic style, and uses this as a condition to drive diffusion models to generate speech that meets user needs; meanwhile, a multi-dimensional validator is designed to self-optimize the generated results.
Embedding Domain Knowledge for Large Language Models via Reinforcement Learning from Augmented Generation
Chaojun Nie (Chinese Academy of Sciences), Zichen Wang (China Southern Power Grid Artificial Intelligence Technology Co., Ltd.)
Explainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextRetrieval-Augmented Generation
🎯 What it does: Proposes a training method for embedding domain knowledge into large language models through reinforcement learning from incremental generation (RLAG).
Emergent morpho-phonological representations in self-supervised speech models
Jon Gauthier (University of California, San Francisco), Edward F. Chang (University of California, San Francisco)
RecognitionRepresentation LearningTransformerContrastive LearningTextAudio
🎯 What it does: Investigate the internal representations used by the self-supervised speech model (S3M) in word recognition tasks, with a particular focus on phonological/morphological phenomena in English plural nouns and third-person singular verbs.
EMNLP: Educator-role Moral and Normative Large Language Models Profiling
Yilin Jiang (Zhejiang University of Technology), Binghao Tu (ZheJiang University)
Safty and PrivacyExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Designed and implemented the EMNLP framework to evaluate ethical risks in teacher-role LLMs under personality traits, moral development stages, and soft prompt injection.
EMO: Embedding Model Distillation via Intra-Model Relation and Optimal Transport Alignments
Minh-Phuc Truong (Hanoi University of Science and Technology), Trung Le (Monash University)
Knowledge DistillationRepresentation LearningTransformerTextBenchmark
🎯 What it does: Designed a cross-tokenizer knowledge distillation framework called EMO, which can compress large text embedding models into smaller, high-quality models.
EmoAgent: Assessing and Safeguarding Human-AI Interaction for Mental Health Safety
Jiahao Qiu (Princeton University), Mengdi Wang (Princeton University)
Safty and PrivacyLarge Language ModelAgentic AITextBiomedical Data
🎯 What it does: Propose the EmoAgent framework, integrating EmoEval to simulate users susceptible to mental health issues and using PHQ-9, PDI, and PANSS to assess emotional changes, combined with EmoGuard for real-time monitoring and intervention, significantly reducing the risk of emotional deterioration caused by role-play conversations.
Emotion Transfer with Enhanced Prototype for Unseen Emotion Recognition in Conversation
Kun Peng (Institute of Information Engineering Chinese Academy of Sciences), Philip S. Yu (University of Illinois at Chicago)
ClassificationRecognitionTransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposes the task of Unseen Emotion Recognition in Conversations (UERC) and designs the ProEmoTrans framework to accomplish this task.
Empowering GraphRAG with Knowledge Filtering and Integration
Kai Guo (Michigan State University), Jiliang Tang (Michigan State University)
RetrievalTransformerLarge Language ModelGraphRetrieval-Augmented Generation
🎯 What it does: Propose and implement the GraphRAG-FI framework, which applies a two-stage filtering mechanism to GraphRAG and integrates it with LLM internal reasoning to improve the accuracy of knowledge graph question answering.
Empowering Math Problem Generation and Reasoning for Large Language Model via Synthetic Data based Continual Learning Framework
Qian Wan (Central China Normal University), Jianwen Sun (Central China Normal University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningAgentic AIText
🎯 What it does: Proposed a continual learning framework based on synthetic data, utilizing two mechanisms: self-play and multi-agent collaboration to generate high-quality synthetic data, and alternately training through supervised fine-tuning (SFT) and direct preference optimization (DPO) to enhance the capabilities of large language models in mathematical problem generation (MPG) and reasoning (MR).
EnAnchored-X2X: English-Anchored Optimization for Many-to-Many Translation
Sen Yang (Nanjing University), Shanbo Cheng (ByteDance Research)
GenerationData SynthesisOptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningContrastive LearningTextBenchmark
🎯 What it does: By leveraging English parallel data and the en2x capabilities of LLMs, this paper proposes an English-anchored x2x translation generation and evaluation framework, generating and filtering synthetic data while performing preference optimization to enhance the performance of LLMs in multilingual x2x translation.
End-to-End Learnable Psychiatric Scale Guided Risky Post Screening for Depression Detection on Social Media
Bichen Wang (Harbin Institute of Technology), Bing Qin (Harbin Institute of Technology)
ClassificationTransformerText
🎯 What it does: Designed and implemented an end-to-end learnable mental health scale-guided risk post screening model (E2-LPS), which simultaneously trains the screening and detection processes in depression detection tasks. The model uses a psychological scale (BDI-II) template to assign risk scores to social media posts and employs a straight-through estimator to achieve differentiable Top-K selection, allowing the filter to directly receive feedback from detection performance.
Enhanced Noun-Noun Compound Interpretation through Textual Enrichment
Bingyang Ye (Brandeis University), James Pustejovsky (Brandeis University)
Explainability and InterpretabilityLarge Language ModelPrompt EngineeringText
🎯 What it does: Propose a text enhancement framework that improves the interpretation of noun-noun compounds by large language models through event-based descriptions and conditional contexts.
Enhancing Chain-of-Thought Reasoning via Neuron Activation Differential Analysis
Yiru Tang (Renmin University of China), Shijin Wang (iFLYTEK AI Research)
Explainability and InterpretabilityTransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper identifies key reasoning neurons by comparing the activation differences of LLM feedforward layer neurons in high-quality and low-quality chain-of-thought reasoning processes, and improves the model's reasoning performance by enhancing or suppressing the activation values of these neurons.
Enhancing Chinese Offensive Language Detection with Homophonic Perturbation
Junqi Wu (South China University Of Technology), Wu Wei (South China University Of Technology)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Constructed a large-scale homophonic Chinese hate speech detection dataset HED-COLD, and proposed a homophone-aware pre-training and fine-tuning framework to enhance the model's robustness against homophonic attacks.
Enhancing Efficiency and Exploration in Reinforcement Learning for LLMs
Mengqi Liao (Beijing Jiaotong University), Huaiyu Wan (Beijing Jiaotong University)
Computational EfficiencyTransformerReinforcement LearningTextBenchmark
🎯 What it does: Propose a dynamic mechanism for allocating rollout budget and a temperature scheduler to enhance the training efficiency and exploration capability of LLMs during the reinforcement learning process.
Enhancing Large Language Model for Knowledge Graph Completion via Structure-Aware Alignment-Tuning
Yu Liu (Institute of Information Engineering, Chinese Academy of Sciences), Shirui Pan (Institute of Computing Technology, Chinese Academy of Sciences)
Representation LearningGraph Neural NetworkTransformerLarge Language ModelContrastive LearningGraph
🎯 What it does: Proposed the SAT framework, which uses structural-aware alignment tuning to enhance the reasoning performance of LLMs in knowledge graph completion tasks
Enhancing Large Vision-Language Models with Ultra-Detailed Image Caption Generation
Yu Zeng (MoE Key Lab of BIPC, USTC), Feng Zhao (Huawei Noah's Ark Lab)
Data SynthesisTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Designed and implemented a complete pipeline from preprocessing to post-processing, generating high-quality, ultra-detailed image captions, and leveraging these data to significantly enhance the performance of large vision-language models.
Enhancing LLM Language Adaption through Cross-lingual In-Context Pre-training
Linjuan Wu (Zhejiang University), Weiming Lu (Zhejiang University)
Domain AdaptationRepresentation LearningTransformerLarge Language ModelPrompt EngineeringTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed and implemented Cross-lingual In-context Pre-training (CrossIC-PT), which enhances the cross-lingual transfer capability of large language models (LLMs) by interleaving semantically related bilingual texts into a single context window through continuous pre-training.
Enhancing LLM Text Detection with Retrieved Contexts and Logits Distribution Consistency
Zhaoheng Huang (Renmin University of China), Zhicheng Dou (Renmin University of China)
ClassificationLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Propose an unsupervised LLM text detection method named HALO, which uses retrieved relevant human texts and their LLM rewritten versions as context, and determines the source of input text by measuring the logit distribution consistency between the input text and the two types of contexts.
Enhancing LLM-Based Social Bot via an Adversarial Learning Framework
Fanqi Kong (Peking University), Xue Feng (State Key Laboratory of General Artificial Intelligence)
GenerationData SynthesisGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningTextGraph
🎯 What it does: EvoBot, a social robot based on a large language model, generates realistic text by supervised fine-tuning on real user data and iterative optimization within an adversarial learning framework that involves a co-adaptive detector.
Enhancing Logical Reasoning in Language Models via Symbolically-Guided Monte Carlo Process Supervision
Xingwei Tan (University of Sheffield), Nikolaos Aletras (University of Sheffield)
Explainability and InterpretabilityComputational EfficiencyLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Proposes symbolic ReAct along with Monte Carlo process supervision to generate high-quality symbolic reasoning trajectories, and uses them to fine-tune LLMs for improved logical reasoning.
Enhancing Reasoning Abilities of Small LLMs with Cognitive Alignment
Wenrui Cai (Shanghai Jiao Tong University), Xiangzhong Fang (Shanghai Jiao Tong University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AITextBenchmarkChain-of-Thought
🎯 What it does: Developed the CRV+CogPO framework, which utilizes multi-agent collaboration to generate cognitively aligned Chain-of-Thought (CoT) data, thereby enhancing the reasoning capabilities of small-scale LLMs.
Enhancing RLHF with Human Gaze Modeling
Karim Galliamov (Innopolis University), Ilya Pershin (Innopolis University)
Reinforcement Learning from Human FeedbackReinforcement LearningTextMultimodality
🎯 What it does: Integrate a human gaze prediction model into the RLHF framework, accelerating dialogue model training through two methods;
Enhancing Speech Large Language Models with Prompt-Aware Mixture of Audio Encoders
Weiqiao Shan (Northeastern University), JingBo Zhu
RecognitionTransformerLarge Language ModelPrompt EngineeringMixture of ExpertsAudio
🎯 What it does: Propose the Prompt-aware Mixture (PaM) method, integrating multiple audio encoders (Whisper, WavLM, Wav2Vec2) with large language models (LLMs), utilizing prompt-based dynamic routing of different experts for feature fusion to achieve multi-task processing in speech LLMs;
Enhancing Study-Level Inference from Clinical Trial Papers via Reinforcement Learning-Based Numeric Reasoning
Massimiliano Pronesti (IBM Research Europe - Ireland), Yufang Hou (IT:U Interdisciplinary Transformation University Austria)
Explainability and InterpretabilityTransformerSupervised Fine-TuningReinforcement LearningTextBiomedical DataBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose a system based on numerical reasoning to extract structured numerical evidence from full-text clinical trials and infer study-level conclusions.
Enrich-on-Graph: Query-Graph Alignment for Complex Reasoning with LLM Enriching
Songze Li (Zhejiang University), Wen Zhang (Zhejiang University)
Computational EfficiencyRepresentation LearningLarge Language ModelPrompt EngineeringGraph
🎯 What it does: Propose the Enrich-on-Graph (EoG) framework, which leverages the prior knowledge of large language models (LLMs) to enrich query alignment on knowledge graphs (KGs), thereby improving the reasoning quality of knowledge graph question answering (KGQA).
Ensembling Prompting Strategies for Zero-Shot Hierarchical Text Classification with Large Language Models
Mingxuan Xia (Zhejiang University), Gang Chen (Zhejiang University)
ClassificationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Proposed HiEPS, an integrated method based on multiple prompting strategies, to enhance the robustness and accuracy of large language models in zero-shot hierarchical text classification.
EQA-RM: A Generative Embodied Reward Model with Test-time Scaling
Yuhang Chen (University of North Carolina at Chapel Hill), Tianlong Chen (University of North Carolina at Chapel Hill)
Safty and PrivacyExplainability and InterpretabilityRobotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelContrastive LearningVideoMultimodalityBenchmark
🎯 What it does: Designed and trained EQA-RM, a generative multimodal reward model for evaluating Embodied Question Answering (EQA) trajectories, and constructed a specialized evaluation benchmark called EQAREWARDBENCH.
EquiBench: Benchmarking Large Language Models’ Reasoning about Program Semantics via Equivalence Checking
Anjiang Wei (Stanford University), Alex Aiken (Stanford University)
Large Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Proposes the EquiBench benchmark for evaluating the reasoning capabilities of large language models (LLMs) in program semantic equivalence checking.
ESC-Judge: A Framework for Comparing Emotional Support Conversational Agents
Navid Madani (University at Buffalo), Rohini Srihari
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Propose the ESC-Judge framework, which conducts automated, theory-driven evaluations using Hill's E-I-A emotional support model dialogue system.
ESGenius: Benchmarking LLMs on Environmental, Social, and Governance (ESG) and Sustainability Knowledge
Chaoyue He (Alibaba-NTU Global e-Sustainability CorpLab), Chunyan Miao (Alibaba-NTU Global e-Sustainability CorpLab)
TransformerLarge Language ModelTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the ESGenius benchmark, comprising 1,136 expert-validated multiple-choice questions (ESGenius-QA) and a corpus of 231 authoritative ESG documents (ESGenius-Corpus), and evaluated the zero-shot and retrieval-augmented (RAG) performance of 50 LLMs based on this benchmark.
Estimating LLM Consistency: A User Baseline vs Surrogate Metrics
Xiaoyuan Wu (Carnegie Mellon University), Lujo Bauer (Carnegie Mellon University)
Computational EfficiencyTransformerLarge Language ModelTextBenchmark
🎯 What it does: Established a human judgment-based LLM consistency benchmark through evaluations of nearly 3,000 users; subsequently compared various existing consistency metrics (e.g., BERTScore, BLEU, ROUGE, USE, Semantic Entropy) and multiple token-level uncertainty measures (probability, log probability, entropy, DLR), and proposed a 16-logit combination method that approximates human consistency scores without generating multiple responses.
EuroGEST: Investigating gender stereotypes in multilingual language models
Jacqueline Rowe (University of Edinburgh), Alexandra Birch (University of Edinburgh)
TextBenchmark
🎯 What it does: Proposed and constructed the EuroGEST multilingual gender stereotype evaluation dataset, assessing the gender bias in 24 multilingual large language models;
Evaluating and Aligning Human Economic Risk Preferences in LLMs
Jiaxin Liu (Hong Kong University of Science and Technology), Kar Yan Tam (Hong Kong University of Science and Technology)
TransformerLarge Language ModelTabularFinance Related
🎯 What it does: Evaluate the economic rationality of large language models (LLMs) in identifying individual risk preferences and decision-making, propose the Risk Disparity Score (RDS) evaluation metric, and compare the performance of three LLMs under different risk scenarios; further adopt Direct Preference Optimization (DPO) and In-Context Learning (ICL) alignment methods to enhance the model's alignment with individual risk preferences.
Evaluating Behavioral Alignment in Conflict Dialogue: A Multi-Dimensional Comparison of LLM Agents and Humans
Deuksin Kwon (University of Southern California), Gale Lucas
TransformerLarge Language ModelText
🎯 What it does: The study evaluates the behavioral consistency between personality-driven LLMs and humans in conflict dialogue scenarios.
Evaluating Cognitive-Behavioral Fixation via Multimodal User Viewing Patterns on Social Media
Yujie Wang (Chinese Academy of Sciences), Jie Zhang (Chinese Academy of Sciences)
Explainability and InterpretabilityRepresentation LearningTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a framework based on multi-modal hierarchical topic extraction and quantification of cognitive fixation behavior, aimed at automatically assessing the cognitive fixation tendencies of social media users.
Evaluating Language Translation Models by Playing Telephone
Syeda Jannatus Saba (Stony Brook University), Steven Skiena (Stony Brook University)
Data SynthesisTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Propose an unsupervised "telephone game" method that generates training data with semantic drift by translating multiple times between the source and target languages, used to train translation evaluation models.
Evaluating Large Language Models for Detecting Antisemitism
Jay Patel (Binghamton University), Jeremy Blackburn (Binghamton University)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: Evaluated the ability of eight open-source large language models to detect antisemitic content and designed Guided-CoT, a prompt method based on 'reasoning steps,' to further quantify differences in model-generated explanations;
Evaluating LLM-Generated Diagrams as Graphs
Chumeng Liang (University of Illinois Urbana-Champaign), Jiaxuan You (University of Illinois Urbana-Champaign)
TransformerLarge Language ModelGraph
🎯 What it does: Proposed DiagramEval, a fine-grained and explainable evaluation framework specifically designed to assess scientific diagrams generated by large language models (LLMs);
Evaluating Robustness of Large Audio Language Models to Audio Injection: An Empirical Study
Guanyu Hou (University Of Electronic Science And Technology Of China), Wenbo Jiang (University Of Electronic Science And Technology Of China)
Adversarial AttackTransformerLarge Language ModelBenchmarkAudio
🎯 What it does: Evaluate the robustness of large-scale audio language models against four types of audio injection attacks (audio interference, instruction following, context injection, judgment hijacking), construct a systematic benchmark, and assess five mainstream models.
Evaluating Spatiotemporal Consistency in Automatically Generated Sewing Instructions
Luisa Geiger (Congree Language Technologies), Alexander Koller (Saarland University)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Designed and validated a tree-based evaluation metric to measure the quality of automatically generated sewing instructions in terms of spatiotemporal consistency;
Evaluating Taxonomy Free Character Role Labeling (TF-CRL) in News Stories using Large Language Models
David G Hobson (McGill University), Andrew Piper (McGill University)
ClassificationTransformerLarge Language ModelText
🎯 What it does: Propose and evaluate a non-categorical role labeling technique (TF-CRL) that generates open-ended, combinable role labels for people in news texts using large language models.
Evaluating the Effectiveness and Scalability of LLM-Based Data Augmentation for Retrieval
Pranjal A Chitale, Amit Sharma (Microsoft Research India)
Data SynthesisRetrievalKnowledge DistillationTransformerLarge Language ModelText
🎯 What it does: This paper conducts systematic incremental experiments on pseudo queries generated by large-scale LLMs, evaluating their performance improvements and scalability limitations for dense retrieval models, using combinations of various incremental scales, densities, and LLM sizes, with over 100 experimental configurations;
Evaluating the Evaluators: Are readability metrics good measures of readability?
Isabel Cachola (Johns Hopkins University), Mark Dredze (Johns Hopkins University)
Explainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper evaluates readability assessment methods for plain language summarization (PLS), comparing the effectiveness of traditional readability metrics with human judgments and large language models (LLMs).
Evaluation and Facilitation of Online Discussions in the LLM Era: A Survey
Katerina Korre (Archimedes, Athena Research Center), John Pavlopoulos
TransformerLarge Language ModelPrompt EngineeringTextReview/Survey Paper
🎯 What it does: Reviews the current status and methods of using large language models (LLM) for quality assessment and promotion of online discussions, and proposes a new classification system for strategies of discussion quality evaluation and promotion;
EverTracer: Hunting Stolen Large Language Models via Stealthy and Robust Probabilistic Fingerprint
Zhenhua Xu (Zhejiang University), Wenpeng Xing (Zhejiang University)
Anomaly DetectionSafty and PrivacyTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Propose EverTracer, which achieves stealthy and robust model fingerprint verification by leveraging gray-box access to LLMs through probabilistic variation signals derived from modified membership inference attacks.
EvolveSearch: An Iterative Self-Evolving Search Agent
Ding-Chu Zhang, Fei Huang (ShanghaiTech University)
TransformerSupervised Fine-TuningReinforcement LearningTextBenchmark
🎯 What it does: Designed and implemented EvolveSearch, a self-evolving framework combining iterative reinforcement learning with supervised fine-tuning (SFT), to enhance large language models (LLMs) in web search and multi-hop question answering tasks.
Evolving Chinese Spelling Correction with Corrector-Verifier Collaboration
Linfeng Liu (Shanghai Jiao Tong University), Hai Zhao (Shanghai Jiao Tong University)
Computational EfficiencyData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes the Automated Error Correction Iteration (ACI) framework, which generates adaptive training data through collaboration between a BERT error correction model and an LLM validator, achieving Chinese spelling error correction improvement without human annotation.
Examining False Positives under Inference Scaling for Mathematical Reasoning
Yu Wang (University of Science and Technology of China), Fuli Feng (University of Science and Technology of China)
Large Language ModelReinforcement LearningTextBenchmarkChain-of-Thought
🎯 What it does: Systematically evaluate the occurrence of 'false positives' in large language models during mathematical reasoning tasks, covering various models, datasets, decoding strategies, and reasoning scaling methods;
ExeCoder: Empowering Large Language Models with Executability Representation for Code Translation
Minghua He (Peking University), Dongmei Zhang (Microsoft)
Representation LearningAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: Designed and implemented ExeCoder, a large language model specifically designed for code translation tasks, enhancing translation accuracy by introducing executable representations.
Expanding before Inferring: Enhancing Factuality in Large Language Models through Premature Layers Interpolation
Dingwei Chen (Sun Yat-Sen University), Chengming Li (Shenzhen MSU-BIT University)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Proposes a training-free, plug-and-play method called 'Premature Layers Interpolation (PLI)', which inserts temporary layers generated via spherical linear interpolation between existing layers in large language models to expand model depth and improve factual accuracy.
ExpandR: Teaching Dense Retrievers Beyond Queries with LLM Guidance
Sijia Yao (Northeastern University), Ge Yu (Tsinghua University)
RetrievalTransformerLarge Language ModelReinforcement LearningContrastive LearningText
🎯 What it does: Enhance retrieval effectiveness by jointly training large language models (LLM) with dense retrievers, leveraging semantically rich query expansions generated by the LLM.
Expectation Preference Optimization: Reliable Preference Estimation for Improving the Reasoning Capability of Large Language Models
Zelin Li (Beijing Institution of Technology), Dawei Song (Beijing Institution of Technology)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: Propose the Expectation Preference Optimization (EPO) algorithm, which improves the reasoning capabilities of large language models (LLMs) in inference tasks by estimating preferences through multiple sampling of response groups;
Explainability and Interpretability of Multilingual Large Language Models: A Survey
Lucas Resck (University of Cambridge), Anna Korhonen (University of Cambridge)
Explainability and InterpretabilityTransformerLarge Language ModelTextReview/Survey Paper
🎯 What it does: This paper reviews the explainability and interpretability research of multilingual large language models (MLLM), systematically classifies and summarizes existing methods, tasks, languages, and resources, and proposes a dedicated explainability evaluation framework for MLLM for the first time;
Explaining Differences Between Model Pairs in Natural Language through Sample Learning
Advaith Malladi (IIIT Hyderabad), Shashank Srivastava (University of North Carolina at Chapel Hill)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelTextTabular
🎯 What it does: Propose the SLED framework, which utilizes gradient optimization to synthesize convergent/divergent samples and delegates them to LLMs to generate natural language explanations for model comparisons, covering both text and structured tasks.
Explicit Learning and the LLM in Machine Translation
Malik Marmonier (Inria), Benoît Sagot (Inria)
GenerationTransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: Utilize encrypted constructed languages and grammar books to investigate the explicit learning capabilities of LLMs.
Exploring Artificial Image Generation for Stance Detection
Zhengkang Zhang (Soochow University), Guodong Zhou (Soochow University)
ClassificationTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes using text-generated artificial images to assist in stance detection. Candidate images are generated via a text-to-image model, and the best image is selected through multi-dimensional evaluation and graph re-ranking. The text and image are then fused for multi-modal stance classification.
Exploring Chain-of-Thought Reasoning for Steerable Pluralistic Alignment
Yunfan Zhang (Columbia University), Smaranda Muresan (Columbia University)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextChain-of-Thought
🎯 What it does: Explored applying Chain-of-Thought (CoT) reasoning methods to achieve steerable pluralistic alignment in large language models, comparing the effectiveness of multiple training strategies.
Exploring Changes in Nation Perception with Nationality-Assigned Personas in LLMs
Mahammed Kamruzzaman (University of South Florida), Gene Louis Kim (University of South Florida)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This study assigns personas with 193 national identities to large language models (LLMs) to investigate their evaluations and perceptions of different countries, and systematically measures the model's biases across regions and countries.
Exploring Large Language Models for Detecting Mental Disorders
Gleb Kuzmin (AIRI), Ivan Smirnov (FRC CSC RAS)
ClassificationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText
🎯 What it does: Compare the effectiveness of traditional machine learning, encoder models, and large language models (LLMs) in Russian depression and anxiety text classification tasks, and propose best practices based on LLMs;
Exploring morphology-aware tokenization: A case study on Spanish language modeling
Alba Táboas García (Universitat Pompeu Fabra), Leo Wanner (Barcelona Supercomputing Center)
Representation LearningData-Centric LearningTransformerLarge Language ModelText
🎯 What it does: This paper proposes combining morphology segmentation with subword tokenization by using MorphAGram to perform semi-supervised morphology segmentation on Spanish. The segmentation results are then used to train a BPE tokenizer, which is applied in the RoBERTa pretraining model to verify its improvements in language modeling and downstream tasks.
Exploring Quality and Diversity in Synthetic Data Generation for Argument Mining
Jianzhu Bao (Harbin Institute of Technology), Ruifeng Xu (Harbin Institute of Technology)
Data SynthesisLarge Language ModelText
🎯 What it does: Propose two synthetic data generation schemes based on LLM: quality-oriented (QOS) and diversity-oriented (DOS), to enhance the training of argument mining models.
Exploring Response Uncertainty in MLLMs: An Empirical Evaluation under Misleading Scenarios
Yunkai Dang (Hong Kong University of Science and Technology), Xuming Hu (Hong Kong University of Science and Technology)
Explainability and InterpretabilityLarge Language ModelSupervised Fine-TuningPrompt EngineeringMultimodalityBenchmark
🎯 What it does: Evaluate the response uncertainty of multimodal large language models when encountering misleading prompts and construct the multimodal uncertainty benchmark MUB.
Exploring the Hidden Capacity of LLMs for One-Step Text Generation
Gleb Mezentsev (AIRI Skoltech), Ivan Oseledets (AIRI Skoltech)
GenerationRepresentation LearningTransformerLarge Language ModelText
🎯 What it does: By learning two 'proto-token' embeddings on a frozen LLM, hundreds of accurate tokens are generated in a single forward pass, demonstrating the potential for parallel multi-token generation in LLMs.
Exploring the Impact of Personality Traits on LLM Bias and Toxicity
Shuo Wang (University of Macau), Derek F. Wong (University of Macau)
Safty and PrivacyExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Explore the impact of HEXACO personality traits on bias and toxicity in LLM-generated text, and verify their adjustability by activating different personality levels in prompts.
Exploring the Limitations of Mamba in COPY and CoT Reasoning
Ruifeng Ren (Renmin University of China), Yong Liu (Renmin University of China)
Computational EfficiencyTransformerTextSequentialBenchmarkChain-of-Thought
🎯 What it does: This study evaluates the expressive power and computational cost of the Mamba model in COPY operations and Chain-of-Thought (CoT) reasoning from both theoretical and experimental perspectives, revealing that it struggles to replicate long sequences under constant size and performs worse than Transformer in long-chain CoT tasks;
Extending Automatic Machine Translation Evaluation to Book-Length Documents
Kuang-Da Wang (National Yang Ming Chiao Tung University), Boris Ginsburg (National Yang Ming Chiao Tung University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Proposes a SEGALE evaluation framework that extends existing sentence-level automatic MT evaluation to long documents and even book-level assessments;
Extracting and Combining Abilities For Building Multi-lingual Ability-enhanced Large Language Models
Zhipeng Chen (Renmin University of China), Ji-Rong Wen (Renmin University of China)
TransformerLarge Language ModelText
🎯 What it does: This paper proposes a multilingual capability extraction and combination method called MAEC, which can extract weights related to specific high-level capabilities (such as mathematical reasoning, scientific reasoning) using capability-related corpora in a single language (English), and then transfer these weights to multilingual LLMs through simple addition and subtraction operations, thereby enhancing their cross-lingual high-level reasoning capabilities.
Extracting Linguistic Information from Large Language Models: Syntactic Relations and Derivational Knowledge
Tsedeniya Kinfe Temesgen (Technische Universität München), Alexander Fraser (Technische Universität München)
TransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: Conducted cross-lingual linguistic knowledge extraction experiments on GPT-4o and LLaMA, designing three diagnostic tasks: syntax role labeling at the sentence level, derivational morphology decomposition at the word level, and in-depth morphological analysis for German and Amharic.
Extractive Fact Decomposition for Interpretable Natural Language Inference in one Forward Pass
Nicholas Popovič (TU Dresden), Michael Färber (TU Dresden)
Explainability and InterpretabilityComputational EfficiencyTransformerText
🎯 What it does: Proposes JEDI, an encoder-only model that can perform atomic fact extraction and explainable natural language reasoning from the original text in a single forward pass;
F²Bench: An Open-ended Fairness Evaluation Benchmark for LLMs with Factuality Considerations
Tian Lan (Inner Mongolia University), Guanglai Gao (Inner Mongolia University)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposes F Bench 2, an open fairness evaluation benchmark for assessing LLMs' performance in fairness and factual accuracy.
F2TEval: Human-Aligned Multi-Dimensional Evaluation for Figure-to-Text Task
Tan Yue (Peking University), Dongyan Zhao (Beijing University of Posts and Telecommunications)
GenerationComputational EfficiencyTransformerMixture of ExpertsVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposed F2TEval, a no-reference chart-text generation quality assessment method based on five-dimensional human evaluation criteria, and constructed the F2TBenchmark dataset.
Facilitating Cognitive Accessibility with LLMs: A Multi-Task Approach to Easy-to-Read Text Generation
François Ledoyen (Université Caen Normandie), Youssef Chahir (Université Caen Normandie)
GenerationTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: Generate readable text compliant with the European Easy‑to‑Read (ETR) guidelines using large language models, and propose a multi-task learning framework to enhance generation quality.
Facilitating Long Context Understanding via Supervised Chain-of-Thought Reasoning
Jingyang Lin (University of Rochester), Jiebo Luo (PAII Inc.)
TransformerLarge Language ModelSupervised Fine-TuningAgentic AITextSequentialFinance RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed a long-context question-answering dataset called LongFinanceQA, and enhanced the performance of lightweight LLMs on long-text reasoning through Supervised Chain-of-Thought (CoT) Reasoning
FacLens: Transferable Probe for Foreseeing Non-Factuality in Fact-Seeking Question Answering of Large Language Models
Yanling Wang (Zhongguancun Laboratory), Ke Xu (Zhongguancun Laboratory)
ClassificationDomain AdaptationLarge Language ModelText
🎯 What it does: Studied how to predict the generation of non-factual answers before large language models (LLMs) respond to factual questions, and proposed a lightweight and transferable model called FacLens.
Fair or Framed? Political Bias in News Articles Generated by LLMs
Junho Yoo (Incheon National University), Youhyun Shin (Incheon National University)
GenerationTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper reveals that models still tend to be left-leaning and may invert the stance of cited content when generating text, by evaluating the political bias of seven LLMs in news generation tasks.
FairGen: Controlling Sensitive Attributes for Fair Generations in Diffusion Models via Adaptive Latent Guidance
Mintong Kang (UIUC), Rashmi Gangadharaiah (AWS AI Labs)
GenerationSupervised Fine-TuningReinforcement LearningPrompt EngineeringDiffusion modelImageTextBenchmark
🎯 What it does: Propose a reasoning-time mechanism called FairGen, which is based on adaptive latent guidance and memory modules, enabling text-to-image diffusion models to generate images according to specified sensitive attribute distributions without compromising image quality; simultaneously create Holistic Bias Evaluation Benchmark (HBE), which covers more domains and complex prompts; and evaluate its performance on multiple diffusion models.
FaithUn: Toward Faithful Forgetting in Language Models by Investigating the Interconnectedness of Knowledge
Nakyeong Yang (Seoul National University), Kyomin Jung (Seoul National University)
Safty and PrivacyTransformerLarge Language ModelTextBenchmark
🎯 What it does: Studies how to achieve 'faithful forgetting' of sensitive or private knowledge in large language models (LLMs), and proposes corresponding evaluation benchmarks and methods.
Fann or Flop: A Multigenre, Multiera Benchmark for Arabic Poetry Understanding in LLMs
Wafa Al Ghallabi (Mohamed bin Zayed University of AI), Rao Muhammad Anwer (Mohamed bin Zayed University of AI)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed Fann or Flop—the first benchmark targeting 12 historical eras, 14 poetry genres, encompassing classical and modern free verse Arabic poetry understanding;
FANS: Formal Answer Selection for LLM Natural Language Math Reasoning Using Lean4
Jiarui Yao (University of Illinois Urbana-Champaign), Tong Zhang (University of Illinois Urbana-Champaign)
Large Language ModelTextBenchmarkChain-of-Thought
🎯 What it does: Propose the FANS framework, applying the Lean4 formal language to answer selection in LLM natural language math reasoning;
FaST: Feature-aware Sampling and Tuning for Personalized Preference Alignment with Limited Data
Thibaut Thonet (NAVER Labs Europe), Marc Dymetman (Independent Researcher)
Recommendation SystemExplainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: Propose a PPALLI framework for personalizing large language models under scenarios where users provide only a small number of preference annotations (less than 100), and release two new datasets, DnD and ELIP, within this framework.
Faster In-Context Learning for LLMs via N-Gram Trie Speculative Decoding
Jinglin Chen (Wuhan University), Ping Wang (Wuhan University)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: Proposed a speculative decoding method based on n-gram Trie, utilizing the overlap between context and model output to accelerate inference in large language models
FB-Bench: A Fine-Grained Multi-Task Benchmark for Evaluating LLMs’ Responsiveness to Human Feedback
Youquan Li (Peking University), Wentao Zhang (Peking University)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper proposes and constructs the FB-Bench benchmark to evaluate the response capability of large language models to human feedback in Chinese multi-turn dialogue scenarios.
Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models
Zihao Li (New Jersey Institute of Technology), Mengnan Du (New Jersey Institute of Technology)
Explainability and InterpretabilityRepresentation LearningTransformerAuto EncoderTextBenchmarkChain-of-Thought
🎯 What it does: Proposed a method combining sparse autoencoder (SAE) and SAE-free direction-oriented approach to directly infer and activate internal states of large language models (LLMs) from standard Chain-of-Thought (CoT), enhancing their reasoning capabilities.
FedMABench: Benchmarking Mobile GUI Agents on Decentralized Heterogeneous User Data
WenHao Wang (Zhejiang University), Yanfeng Wang (Shanghai AI Laboratory)
Federated LearningBenchmark
🎯 What it does: Developed and released FedMABench, the first unified benchmark for distributed mobile GUI agents, to evaluate the effectiveness of federated learning training.
Few-Shot Learning Translation from New Languages
Carlos Mullov (Karlsruhe Institute of Technology), Alexander Waibel (Karlsruhe Institute of Technology)
Representation LearningMeta LearningTransformerSupervised Fine-TuningText
🎯 What it does: The study utilizes a small amount of parallel corpus and monolingual corpus in low-resource languages to achieve zero/few-shot machine translation through training high-quality word vectors.
Few-Shot Open-Set Classification via Reasoning-Aware Decomposition
Avyav Kumar Singh (King's College London), Helen Yannakoudakis (King's College London)
ClassificationTransformerLarge Language ModelReinforcement LearningFlow-based ModelTextChain-of-Thought
🎯 What it does: Proposes a reasoning-aware decomposition method based on Wasserstein-Generative Flow Network (W-GFN) for few-shot open-set classification in small-scale LLMs.
FilBench: Can LLMs Understand and Generate Filipino?
Lester James Validad Miranda (Allen Institute for AI), Joseph Marvin Imperial (SEACrowd)
ClassificationGenerationTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed the FILBENCH benchmark to evaluate the understanding and generation capabilities of LLMs in Philippine languages (Filipino, Tagalog, Cebuano).
FillerSpeech: Towards Human-Like Text-to-Speech Synthesis with Filler Insertion and Filler Style Control
Seung-Bin Kim (Korea University), Seong-Whan Lee (Samsung Research)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningFlow-based ModelTextAudio
🎯 What it does: Proposes the FillerSpeech framework to achieve natural filler word insertion and controllable style control.
Financial Risk Relation Identification through Dual-view Adaptation
Wei-Ning Chiu (National Taiwan University), Chuan-Ju Wang (Academia Sinica)
RetrievalRepresentation LearningGraph Neural NetworkTransformerContrastive LearningTextTabularBenchmarkFinance Related
🎯 What it does: Training a specialized financial text retrieval encoder through unsupervised dual-perspective contrastive learning on Form 10-K disclosure documents, and proposing Risk Relationship Score (RRS) based on this encoder to quantify and explain cross-company risk associations.
Finding your MUSE: Mining Unexpected Solutions Engine
Nir Sweed (Hebrew University of Jerusalem), Dafna Shahaf (Hebrew University of Jerusalem)
GenerationTransformerLarge Language ModelTextGraphRetrieval-Augmented Generation
🎯 What it does: Constructed a large-scale Functional Concept Graph (FCG) and proposed the MUSE algorithm to generate creative inspirations on this graph, helping users break free from design rigidity.
Finetuning LLMs for Human Behavior Prediction in Social Science Experiments
Akaash Kolluri (Stanford University), Michael S. Bernstein (Stanford University)
Data-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper constructs a general predictive model capable of simulating human behavior by performing fine-grained labeling on large-scale social science experimental data and subsequently fine-tuning LLMs;
Fingerprinting LLMs through Survey Item Factor Correlation: A Case Study on Humor Style Questionnaire
Simon Münker (Trier University)
Explainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: Researchers applied the Humor Styles Questionnaire (HSQ) to large language models (LLMs), recorded their responses to 32 items, constructed factor correlation matrices for each model, and formed so-called 'fingerprints' to investigate the internal representation of psychological constructs in LLMs.
FinMTEB: Finance Massive Text Embedding Benchmark
Yixuan Tang (Hong Kong University of Science and Technology), Yi Yang (Hong Kong University of Science and Technology)
ClassificationData SynthesisRetrievalTransformerLarge Language ModelContrastive LearningTextBenchmarkFinance Related
🎯 What it does: Proposed a novel text embedding benchmark for the financial domain, FinMTEB, and developed FinE5, a financial-adapted embedding model fine-tuned using persona-based synthetic data.