Papers by Wen Wu

Towards Self-Evolving Agents: Enabling Autonomy through Interactive Experience Refinement (2026.findings-acl)

Challenge:	Temporal knowledge graphs (TKGs) require predicting future facts by modeling structural dependencies within each snapshot and temporal evolution across snapshots.
Approach:	They propose an encoder-agnostic framework that provides persistent entity states . EST maintains a global state buffer and aligns structural evidence with sequential signals .
Outcome:	Experiments show that EST improves diverse backbones and achieves state-of-the-art performance.

Copied to clipboard

Cheng Yang, Xuemeng Yang, Licheng Wen, Daocheng Fu, Jianbiao Mei, Rong Wu, Pinlong Cai, Yufan Shen, Nianchen Deng, Jia Xu, Botian Shi, Yu Qiao, Haifeng Li

Challenge:	Large Language Models struggle with complex, multi-step operational tasks because they remain static during inference and cannot learn from past experience.
Approach:	They propose a framework that organizes cross-domain insights to facilitate orchestration of long-horizon workflows.
Outcome:	The proposed framework outperforms existing methods on the TAC productivity benchmark and shows strong cross-task transferability.

From Selection to Refinement: Iterative Optimization for Instruction Data (2026.acl-long)

Copied to clipboard

Hang Hu, Ziyan Liu, Rujie Wen, Ruihui Hou, Xueyan Wu, Mu Zhang, Jianxing Yu, Tong Ruan, Jingping Liu

Estimating the Uncertainty in Emotion Attributes using Deep Evidential Regression (2023.acl-long)

Challenge:	Existing methods to optimize instruction tuning datasets face two main challenges: unreasonable pruning of potentially valuable low-quality data and the persistence of noise or semantic drift during revision.
Approach:	They propose an automated iterative framework for instruction data optimization that prunes low-quality data and refines low quality data using feedback-driven iteration.
Outcome:	The proposed framework outperforms state-of-the-art methods on seven public benchmark datasets with high data efficiency.

Copied to clipboard

Wen Wu, Chao Zhang, Philip Woodland

Sentiment Analysis

Challenge:	Existing methods to predict human emotions are inconsistent due to complexity of emotion and subjectivity of perception.
Approach:	They propose a Bayesian approach to estimate uncertainty in emotion attributes using a deep neural network model.
Outcome:	The proposed approach estimates uncertainty in emotion attributes along with aleatoric and epistemic uncertainties.

Masked Thought: Simply Masking Partial Reasoning Steps Can Improve Mathematical Reasoning Learning of Language Models (2024.acl-long)

Copied to clipboard

Changyu Chen, Xiting Wang, Ting-En Lin, Ang Lv, Yuchuan Wu, Xin Gao, Ji-Rong Wen, Rui Yan, Yongbin Li

CORAL: Benchmarking Multi-turn Conversational Retrieval-Augmented Generation (2025.findings-naacl)

Challenge:	Despite the advances in large language models, they still face difficulties with multi-step reasoning tasks.
Approach:	They propose a method that randomly masks certain tokens within the chain of thought to improve model accuracy by 5% over standard supervised fine-tuning.
Outcome:	The proposed method improves accuracy and accuracy by 5% over standard fine-tuning with a few codes modified.

Copied to clipboard

Yiruo Cheng, Kelong Mao, Ziliang Zhao, Guanting Dong, Hongjin Qian, Yongkang Wu, Tetsuya Sakai, Ji-Rong Wen, Zhicheng Dou

Natural Language Interfaces

Challenge:	Existing research focuses on single-turn RAG, leaving a gap in addressing multi-turn conversations . a new benchmark is designed to assess RAG systems in realistic multi-turned conversations based on Wikipedia .
Approach:	They propose a large-scale benchmark to assess RAG systems in multi-turn contexts . CORAL includes diverse information-seeking conversations automatically derived from Wikipedia . authors propose unified framework to standardize various conversational RAG methods .
Outcome:	The proposed framework supports three core tasks of conversational RAG: passage retrieval, response generation, and citation labeling.

Calibration-Aware Policy Optimization for Reasoning LLMs (2026.acl-long)

Copied to clipboard

Ziqi Wang, Xingzhou Lou, Meiqi Wu, Zhengqi Wen, Junge Zhang

COM2SENSE: A Commonsense Reasoning Benchmark with Complementary Sentences (2021.findings-acl)

Challenge:	Existing approaches to model calibration are limited or sacrifice gains in reasoning accuracy.
Approach:	They propose a method that improves calibration by 15% while boosting accuracy by 5% . they propose GRPO-style algorithms that misalign uncertainty-agnostic advantage estimation .
Outcome:	The proposed approach improves calibration by 15% while achieving comparable to or better than GRPO on multiple mathematical reasoning benchmarks.

Copied to clipboard

Shikhar Singh, Nuan Wen, Yu Hou, Pegah Alipoormolabashi, Te-lin Wu, Xuezhe Ma, Nanyun Peng

Defending against Indirect Prompt Injection by Instruction Detection (2025.findings-emnlp)

Challenge:	Recent advances in pretrained language models have shown promising results on commonsense reasoning benchmark datasets.
Approach:	They propose a commonsense reasoning benchmark dataset with 4k sentence pairs . they propose 'gamified' model-in-the-loop setup to incentivize challenging samples .
Outcome:	The proposed benchmarks show that the proposed model achieves 71% standard accuracy and 51% pairwise accuracy, well below human performance.

Copied to clipboard

Tongyu Wen, Chenglong Wang, Xiyuan Yang, Haoyu Tang, Yueqi Xie, Lingjuan Lyu, Zhicheng Dou, Fangzhao Wu

ZoFia: Zero-Shot Fake News Detection with Entity-Guided Retrieval and Multi-LLM Interaction (2026.findings-acl)

Challenge:	Indirect Prompt Injection attacks can be exploited by LLMs that are embedded with external data.
Approach:	They propose a detection-based approach that leverages the behavioral states of LLMs to identify potential IPI attacks.
Outcome:	The proposed approach reduces the success rate of attacks to 0.03% on the BIPIA benchmark.

Copied to clipboard

Lvhua Wu, Xuefeng Jiang, Sheng Sun, Yan Lei, Tian Wen, Yuwei Wang, Min Liu

TOME: A Two-stage Approach for Model-based Retrieval (2023.acl-long)

Challenge:	Large language models (LLMs) are limited by knowledge cutoff and can generate factual hallucinations when handling time-sensitive news.
Approach:	They propose a two-stage zero-shot fake news detection framework that uses a hierarchical salience and saliency-calibrated minimum margin of relevance algorithm to extract core entities accurately.
Outcome:	The proposed framework outperforms existing zero-shot baselines and even most few-shot methods on two public datasets.

Copied to clipboard

Ruiyang Ren, Wayne Xin Zhao, Jing Liu, Hua Wu, Ji-Rong Wen, Haifeng Wang

CHEF: A Pilot Chinese Dataset for Evidence-Based Fact-Checking (2022.naacl-main)

Challenge:	Recent research has focused on model-based retrieval, which discards the index in the traditional retrieval model and memorizes the candidate corpora using model parameters.
Approach:	They propose a model-based retrieval approach that discards the index in the traditional retrieval model and memorizes the candidate corpora using model parameters.
Outcome:	The proposed approach eliminates the index in the traditional retrieval model and memorizes the candidate corpora using model parameters.

Copied to clipboard

Xuming Hu, Zhijiang Guo, GuanYu Wu, Aiwei Liu, Lijie Wen, Philip Yu

Enhancing Automated Essay Scoring Performance via Fine-tuning Pre-trained Language Models with Combination of Regression and Ranking (2020.findings-emnlp)

Challenge:	CHEF dataset provides evidence retrieval over non-English claims . e-fact-checking is a time-consuming task, which can take journalists several hours or days.
Approach:	They construct a dataset of 10K real-world claims that is based on annotated evidence retrieved from the Internet.
Outcome:	The proposed dataset provides evidence retrieval as a latent variable and can be used to train and reason over non-English claims.

Copied to clipboard

Ruosong Yang, Jiannong Cao, Zhiyuan Wen, Youzheng Wu, Xiaodong He

Symmetrical Visual Contrastive Optimization: Aligning Vision-Language Models with Minimal Contrastive Images (2025.acl-long)

Challenge:	Recent work on sentence prediction tasks uses shallow neural networks to learn essay representations and constrain calculated scores with regression loss or ranking loss.
Approach:	They propose to use a pre-trained language model to learn text representations first and then to constrain the scores with regression loss or ranking loss.
Outcome:	The proposed model outperforms state-of-the-art models on the Automated Student Assessment Prize dataset.

Copied to clipboard

Shengguang Wu, Fan-Yun Sun, Kaiyue Wen, Nick Haber

Two-Stage Regularization-Based Structured Pruning for LLMs (2026.acl-long)

Challenge:	Recent studies have shown that Large Vision-Language Models (VLMs) tend to neglect image content and over-rely on language-model priors, resulting in errors in visually grounded tasks and hallucinations.
Approach:	They propose a novel finetuning objective that steers the model toward capturing important visual details and aligning them with corresponding text tokens.
Outcome:	The proposed method achieves up to 22% reduction in hallucinations and significant gains in vision-centric and general tasks while maintaining or improving the model's general abilities.

Copied to clipboard

Mingkuan Feng, Jinyang Wu, Siyuan Liu, Shuai Zhang, Hongjian Fang, Ruihan Jin, Feihu Che, Pengpeng Shao, Zhengqi Wen, Jianhua Tao

FAITH: Factuality Alignment through Integrating Trustworthiness and Honestness (2026.findings-acl)

Challenge:	Structural pruning is a promising solution for large language models . prior structured pruning methods remove unimportant parameters based on certain metrics .
Approach:	They propose a structural pruning method that iteratively learns the weights of transformer layers by adding their l1-norm to the loss function.
Outcome:	The proposed pruning method outperforms strong layer-wise pruning methods without requiring retraining.

Copied to clipboard

Xiaoning Dong, Chengyan Wu, Yajie Wen, Yu Chen, Yun Xue, Zhang Jing, Wei Xu, Bolei Ma

Multi-Programming Language Sandbox for LLMs (2025.acl-demo)

Challenge:	Existing approaches to correct factually inaccurate outputs are lacking the semantic richness needed to properly understand its internal states of trustworthiness and honesty.
Approach:	They propose a framework for factuality alignment that integrates natural-language uncertainty signals with external knowledge and computes confidence scores and semantic entropy from LLM outputs.
Outcome:	Extensive experiments on four knowledge-intensive benchmarks show that FAITH improves the factual accuracy and truthfulness of Large Language Models (LLMs).

Copied to clipboard

Shihan Dou, Jiazheng Zhang, Jianxiang Zang, Yunbo Tao, Weikang Zhou, Haoxiang Jia, Shichun Liu, Yuming Yang, Shenxi Wu, Zhiheng Xi, Muling Wu, Rui Zheng, Changze Lv, Limao Xiong, Shaoqing Zhang, Lin Zhang, Wenyu Zhan, Rongxiang Weng, Jingang Wang, Xunliang Cai, Yueming Wu, Ming Wen, Yixin Cao, Tao Gui, Xipeng Qiu, Qi Zhang, Xuanjing Huang

SPARK: Strategic Policy-Aware Exploration via Dynamic Branching for Long-Horizon Agentic Learning (2026.acl-long)

Challenge:	MPLSandbox is an out-of-the-box multi-programming language sandbox designed to provide unified and comprehensive feedback from compiler and analysis tools for Large Language Models (LLMs).
Approach:	They propose a multi-programming language sandbox that provides unified feedback from compilers and analysis tools for Large Language Models.
Outcome:	The proposed multi-language sandbox can provide comprehensive feedback from compilers and analysis tools for large language models (LLMs).

Copied to clipboard

Jinyang Wu, Shuo Yang, Yuhao Shen, Shuai Zhang, Zhengqi Wen, Jianhua Tao

Challenge:	Existing methods for training large language models waste computation budget on trivial steps while failing to guarantee sample quality.
Approach:	They propose a framework that selectively branches at critical decision states for resource-efficient exploration.
Outcome:	The proposed framework activates adaptive branching exploration at critical decision states to probe promising trajectories, thereby achieving precise resource allocation that prioritizes sampling quality over blind coverage.

TrendSim: Simulating Trending Topics in Social Media Under Poisoning Attacks with LLM-based Multi-agent System (2025.findings-naacl)

Copied to clipboard

Zeyu Zhang, Jianxun Lian, Chen Ma, Yaning Qu, Ye Luo, Lei Wang, Rui Li, Xu Chen, Yankai Lin, Le Wu, Xing Xie, Ji-Rong Wen

Handling Ambiguity in Emotion: From Out-of-Domain Detection to Distribution Estimation (2024.acl-long)

Challenge:	Trending topics bring in a new channel for poisoning attacks, resulting in negative impacts on society.
Approach:	They propose an LLM-based multi-agent system to simulate trending topics in social media . they propose a time-aware interaction mechanism, centralized message dissemination, and an interactive system .
Outcome:	The proposed system simulates trending topics under poisoning attacks on social media platforms.

Copied to clipboard

Wen Wu, Bo Li, Chao Zhang, Chung-Cheng Chiu, Qiujia Li, Junwen Bai, Tara Sainath, Phil Woodland

Virtual Data Augmentation: A Robust and General Framework for Fine-tuning Pre-trained Models (2021.emnlp-main)

Challenge:	Experimental results show that incorporating utterances without majority-agreed labels into an additional class reduces the classification performance of the other emotion classes.
Approach:	They propose to combine utterances without majority-agreed labels into an additional class . they propose to quantify uncertainty in emotion classification using evidential deep learning .
Outcome:	The proposed method retains classification accuracy while effectively detects ambiguous emotion expressions.

Copied to clipboard

Kun Zhou, Wayne Xin Zhao, Sirui Wang, Fuzheng Zhang, Wei Wu, Ji-Rong Wen

A Multi-Modal Knowledge Graph for Classical Chinese Poetry (2022.findings-emnlp)

Challenge:	Recent studies have shown that powerful pre-trained language models can be fooled by small perturbations or intentional attacks.
Approach:	They propose a framework for fine-tuning PLMs using a masked language model and Gaussian noise to augment semantically relevant examples with sufficient diversity.
Outcome:	The proposed framework improves the robustness of pre-trained language models and alleviates performance degradation under adversarial attacks.

Copied to clipboard

Yuqing Li, Yuxin Zhang, Bin Wu, Ji-Rong Wen, Ruihua Song, Ting Bai

Beyond Hard Masks: Progressive Token Evolution for Diffusion Language Models (2026.acl-long)

Challenge:	Existing studies in classical Chinese poetry area focus on generation and analysis of poetry.
Approach:	They propose to integrate the visual information of words in classical Chinese poetry into a multi-modal knowledge graph.
Outcome:	The proposed model bridges the semantic gap between two modalities and achieves state-of-the-art performance on the poetry-image retrieval task.

Copied to clipboard

Linhao Zhong, Linyu Wu, Bozhen Fang, Tianjian Feng, Chenchen Jing, Wen Wang, Jiaheng Zhang, Hao Chen, Chunhua Shen

PAIR: Leveraging Passage-Centric Similarity Relation for Improving Dense Passage Retrieval (2021.findings-acl)

Challenge:	Existing Diffusion Language Models rely on hard binary masking and discrete token assignments, which hinder the revision of early decisions.
Approach:	They propose a diffusion-based language modeling approach that replaces hard binary masks with evolving soft token distributions.
Outcome:	The proposed approach outperforms existing DLMs on multiple benchmarks.

Copied to clipboard

Ruiyang Ren, Shangwen Lv, Yingqi Qu, Jing Liu, Wayne Xin Zhao, QiaoQiao She, Hua Wu, Haifeng Wang, Ji-Rong Wen

Breaking the Evaluation Paradox: Evaluating High-Entropy Search with Computationally Irreducible Constraints (2026.findings-acl)

Challenge:	Recent studies only consider query-centric similarity relation when learning the dual-encoder retriever.
Approach:	They propose a query-centric and PAssage-centric approach to capture more comprehensive similarity relations for dense passage retrieval.
Outcome:	The proposed approach significantly outperforms existing models on both MSMARCO and Natural Questions datasets.

Copied to clipboard

Juntao Wu, Wei Wen, Xianting Huang, Shuai Pang, Ruizhi Qiao, Xing Sun, Ke Wang

Efficient Self-Evaluation for Diffusion Language Models via Sequence Regeneration (2026.acl-long)

Challenge:	a new framework for evaluation of exhaustive search capabilities is needed . high-entropy enumeration tasks make such ground truth impossible for humans to create . VERITAS is a framework built on the principle of computationally irreducible constraints .
Approach:	They propose a framework that uses non-optimizable constraints to create verifiable searches . VERITAS can generate infinite number of test cases with perfect ground truth and precise difficulty control .
Outcome:	a new evaluation framework for large language models is based on non-optimizable constraints . the framework can generate infinite number of test cases with perfect ground truth and precise difficulty control .

Copied to clipboard

Linhao Zhong, Linyu Wu, Wen Wang, Yuling Xi, Chenchen Jing, Jiaheng Zhang, Hao Chen, Chunhua Shen

OpenS2S: Advancing Fully Open-Source End-to-End Empathetic Large Speech Language Model (2025.emnlp-demos)

Challenge:	Non-sequential and bidirectional nature of diffusion large language models makes direct likelihood-based self-evaluation challenging.
Approach:	They propose a self-evaluation confidence quantification method for diffusion large language models that quantifies confidence by computing the probability of regenerating tokens in the entire generated sequence, given the full context.
Outcome:	The proposed method is correlated with semantic coherence and answer accuracy.

Copied to clipboard

Chen Wang, Tianyu Peng, Wen Yang, YiNan Bai, Guangfu Wang, Jun Lin, Lanpeng Jia, Lingxiang Wu, Jinqiao Wang, Chengqing Zong, Jiajun Zhang

RadialRouter: Structured Representation for Efficient and Robust Large Language Models Routing (2025.findings-emnlp)

Challenge:	Empathetic speech models are increasingly closed off, leaving details about the architecture, data and development opaque to researchers.
Approach:	They propose an open-source empathetic speech-to-text model with a streaming interleaved decoding architecture and a data pipeline to enable end-to end training.
Outcome:	The proposed model is open-source and transparent, with no data or data required to build it.

Copied to clipboard

Ruihan Jin, Pengpeng Shao, Zhengqi Wen, Jinyang Wu, Mingkuan Feng, Shuai Zhang, Jianhua Tao

Glance-or-Gaze: Incentivizing LMMs to Adaptively Focus Search via Reinforcement Learning (2026.findings-acl)

Challenge:	Current routing methods are limited in exploring the connection between query and LLM characteristics.
Approach:	They propose a framework for LLM routing that uses a transformer-based backbone and a radial structure to articulate the query-LLMs relationship.
Outcome:	The proposed framework outperforms existing routing methods by 9.2% and 5.8% on RouterBench.

Copied to clipboard

Hongbo Bai, Yujin Zhou, Yile Wu, Chi-Min Chan, Pengcheng Wen, Kunhao Pan, Sirui Han, Yike Guo

InferPilot: Autonomous Inference Attacks Against ML Services With LLM-Based Agents (2026.findings-acl)

Challenge:	Existing search-augmented approaches rely on indiscriminate whole-image retrieval and lack deep iterative reflection, limiting their effectiveness on complex visual queries.
Approach:	They propose a fully autonomous framework that shifts from passive perception to active visual planning and introduces a Selective Gaze mechanism that dynamically chooses whether to glance at global context or gaze into high-value regions.
Outcome:	Experiments across six benchmarks demonstrate state-of-the-art performance.

Copied to clipboard

Yixin Wu, Rui Wen, Chi Cui, Michael Backes, Yang Zhang

Learning Intrinsic Dimension via Information Bottleneck for Explainable Aspect-based Sentiment Analysis (2024.lrec-main)

Challenge:	Inference attacks are important for assessing model's robustness, but their implementation and parameters are challenging for non-experts.
Approach:	They propose an autonomous agent capable of conducting inference attacks without human intervention.
Outcome:	The proposed agent achieves a 100.0% task completion rate and near-expert attack performance with an average token cost of only 0.627 per run.

Copied to clipboard

Zhenxiao Cheng, Jie Zhou, Wen Wu, Qin Chen, Liang He

TemplateRL: Structured Template-Guided Reinforcement Learning for LLM Reasoning (2026.findings-acl)

Challenge:	Gradient-based explanation methods are increasingly used to interpret neural models in natural language processing (NLP) however, in the context of Aspect-based Sentiment Analysis, only specific dimensions are pertinent.
Approach:	They propose a Gradient-based explanation framework that leverages an information bottleneck to refine word embeddings into a concise intrinsic dimension, maintaining essential features and omitting unrelated information.
Outcome:	The proposed framework improves both the models’ performance and explanations’ clarity by identifying sentiment-aware features.

Copied to clipboard

Jinyang Wu, Chonghua Liao, Mingkuan Feng, Shuai Zhang, Zhengqi Wen, Haoran Luo, Ling Yang, Huazhe Xu, Jianhua Tao

CHESS: Optimizing LLM Inference via Channel-Wise Thresholding and Selective Sparsification (2024.emnlp-main)

Challenge:	Existing RL methods rely on unstructured self-sampling to fit scalar rewards, resulting in inefficient rollouts.
Approach:	They propose a structured template-guided RL framework that augments policy optimization with explicit template guidance.
Outcome:	Experiments show that TemplateRL outperforms GRPO and GRPI by 99% on AIME and 41% on AMC with superior stability on weak models and remarkable cross-domain generalization.

Copied to clipboard

Junhui He, Shangyu Wu, Weidong Wen, Chun Jason Xue, Qingan Li

ImF: Embedding an Implicit Fingerprint in Your Large Language Models (2026.acl-long)

Challenge:	Existing methods for activation sparsification do not capture the relationship between activation and model performance.
Approach:	They propose a general activation sparsification approach using channel-wise thresholding and selective sparsifying to capture the relationship between activation and model performance.
Outcome:	The proposed approach reduces the number of activated neurons during inference by 1.27x over eight downstream tasks while activating fewer parameters than existing methods.

Copied to clipboard

Jiaxuan Wu, Wanli Peng, Hang fu, Xue Yiming, Juan Wen

Smart-Searcher: Incentivizing the Dynamic Knowledge Acquisition of LLMs via Reinforcement Learning (2025.findings-emnlp)

Challenge:	Training and serving large language models (LLMs) is resource-intensive, making reliable intellectual property protection and black-box ownership verification increasingly important.
Approach:	They propose a method to inject a small set of secret query–response behaviors into model fingerprinting . they encode ownership information into a natural-looking target response and derive a semantically aligned query .
Outcome:	The proposed fingerprints improve stealthiness and remain verifiable under model updates and deployment-time prompt interventions.

Copied to clipboard

Huatong Song, Jinhao Jiang, Wenqing Tian, Zhipeng Chen, Yuhuan Wu, Jiahao Zhao, Yingqian Min, Xin Zhao, Lei Fang, Ji-Rong Wen

LongWeave: A Long-Form Generation Benchmark Bridging Real-World Relevance and Verifiability (2025.findings-emnlp)

Challenge:	Large Language Models (LLMs) are powerful but prone to hallucinations due to static knowledge. Retrieval-augmented generation (RAG) helps by injecting external information, but current methods are costly, generalize poorly, or ignore the model’s internal knowledge.
Approach:	They propose a framework to train large language models to leverage both internal and external knowledge sources.
Outcome:	The proposed framework outperforms existing methods and achieves efficient retrieval-augmented reasoning.

Copied to clipboard

Zikai Xiao, Fei Huang, Jianhong Tu, Jianhui Wei, Wen Ma, Yuxuan Zhou, Jian Wu, Bowen Yu, Zuozhu Liu, Junyang Lin

Challenge:	Existing benchmarks for long-form generation assess real-world queries with hard-to-verify metrics or use synthetic setups that overlook real-life intricacies.
Approach:	They propose a new approach that balances verifiable and real-world assessment with Target-Anchored Evaluation.
Outcome:	The proposed model balances real-world and verifiable assessment with Target-Anchored Evaluation (TAE) it generates queries, textual materials, and anchors based on verifier targets within real-life scenarios .

Filter-And-Refine: A MLLM Based Cascade System for Industrial-Scale Video Content Moderation (2025.acl-industry)

Copied to clipboard

Zixuan Wang, Jinghao Shi, Hanzhong Liang, Xiang Shen, Vera Wen, Zhiqian Chen, Yifan Wu, Zhixin Zhang, Hongyu Xiong

LLaDA 1.5: Variance-Reduced Preference Optimization for Large Language Diffusion Models (2026.acl-long)

Challenge:	Effective content moderation is essential for video platforms to safeguard user experience and uphold community standards.
Approach:	They propose a method to transform a generative MLLM into a multimodal classifier using minimal discriminative training data.
Outcome:	The proposed method improves F1 score by 66.50% over traditional classifiers while requiring only 2% of the fine-tuning data.

Copied to clipboard

Fengqi Zhu, Rongzhen Wang, Shen Nie, Xiaolu Zhang, Chunwei Wu, Jun Zhou, Yankai Lin, Ji-Rong Wen, Chongxuan Li

RLSeek: Evidence-Grounded Reasoning for RAG Hallucination Detection (2026.acl-long)

Challenge:	Masked diffusion language models have achieved significant progress in language modeling . however, the systematic analysis and empirical validation of their alignment on general tasks remains underexplored.
Approach:	They propose a framework that analyzes the bias and variance of preference optimization loss and gradient based on Direct Preference Optimization.
Outcome:	The proposed model outperforms its SFT-only predecessor on general benchmarks . it consistently outperformed other strong language models and ARMs on general tasks .

Copied to clipboard

Zhaoheng Huang, Dacheng Wen, Yutao Zhu, Xiaoying Lian, Yushi Liang, Kai Hao, Nan Li, Liangjie Zhang, Qi Zhang, Ji-Rong Wen, Zhicheng Dou, Fangzhao Wu

SLIP: Soft Label Mechanism and Key-Extraction-Guided CoT-based Defense Against Instruction Backdoor in APIs (2026.findings-acl)

Challenge:	Recent work addresses this problem by training span-level hallucination detectors using reinforcement learning and chain-of-thought reasoning.
Approach:	They propose a framework that explicitly enforces active evidence seeking during CoT reasoning by requiring quotation of relevant source segments at each verification step.
Outcome:	The proposed framework improves hallucination span detection performance with limited reasoning overhead and improved robustness in out-of-domain settings.

Copied to clipboard

Wu Zhengxian, Juan Wen, Wanli Peng, Haowei Chang, Yinghan Zhou, Xue Yiming

QiMeng-Attention: SOTA Attention Operator is generated by SOTA Attention Algorithm (2025.findings-acl)

Challenge:	Existing black-box instruction backdoors can detect poisoned inputs, but fail to recover correct outputs once the backdoor is activated.
Approach:	They propose a soft label mechanism and key-extraction-guided CoT-based defense against instruction backdoors in APIs (SLIP) they propose KCOT-based model to extract task-relevant keywords and phrases rather than only considering the single trigger or overall text semantics.
Outcome:	The proposed model reduces the average attack success rate to 25.13% and improves clean accuracy to 87.15% and outperforms state-of-the-art black-box defenses.

Copied to clipboard

Qirui Zhou, Shaohui Peng, Weiqiang Xiong, Haixin Chen, Yuanbo Wen, Haochen Li, Ling Li, Qi Guo, Yongwei Zhao, Ke Gao, Ruizhi Chen, Yanjun Wu, Zhao Chen, Yunji Chen

Programming Languages in NLP

Challenge:	Existing LLMs cannot comprehend the complex data flow and computation process of the attention operator and utilize low-level primitive to exploit GPU performance.
Approach:	They propose an LLM-friendly Thinking Language (LLM-TL) that can decouple the generation of high-level optimization logic and low-level implementation on GPU and enhance LLMs’ understanding of attention operator.
Outcome:	The proposed method outshines existing LLMs on A100, RTX8000, and T4 GPUs, achieving a speed-up of up to 35.16.

Beyond Examples: Towards Automated Thought-level In-Context Reasoning for Large Language Models (2026.acl-long)

Copied to clipboard

Jinyang Wu, Mingkuan Feng, Shuai Zhang, Feihu Che, Zhengqi Wen, Chonghua Liao, Ling Yang, Haoran Luo, Zheng Lian, Jianhua Tao

Peering Behind the Shield: Guardrail Identification in Large Language Models (2026.findings-acl)

Challenge:	In-context learning (ICL) struggles with complex reasoning due to superficial, example-level implicit imitation.
Approach:	They propose an automated method that shifts from surface-level examples to more guidance-oriented thought patterns.
Outcome:	The proposed method achieves 80.6% accuracy on MATH and 62.5% on AMC, surpassing GPT-4o’s 77.2% and 57.5% accuracy.

Copied to clipboard

Ziqing Yang, Yixin Wu, Rui Wen, Michael Backes, Yang Zhang

Investigating the Factual Knowledge Boundary of Large Language Models with Retrieval Augmentation (2025.coling-main)

Challenge:	Identifying guardrails in conversational AI agents is critical for identifying malicious content . identifying guardrail components in black-box AI agents poses security challenges .
Approach:	They propose a method that leverages guard-specific adversarial prompts to detect guardrails in black-box AI agents.
Outcome:	The proposed method achieves perfect classification accuracy in multiple scenarios.

Copied to clipboard

Ruiyang Ren, Yuhao Wang, Yingqi Qu, Wayne Xin Zhao, Jing Liu, Hua Wu, Ji-Rong Wen, Haifeng Wang

RocketQAv2: A Joint Training Method for Dense Passage Retrieval and Passage Re-ranking (2021.emnlp-main)

Challenge:	Large language models (LLMs) have shown impressive prowess in solving a wide range of tasks with world knowledge, but it remains unclear how well they perceive their factual knowledge boundaries.
Approach:	They propose to use a retrieval augmentation approach to enhance LLMs' awareness of factual knowledge boundaries to analyze factual and factual information in open-domain question answering (QA)
Outcome:	The proposed method improves LLMs’ QA and judgemental capabilities by integrating supporting documents with the questions.

Copied to clipboard

Ruiyang Ren, Yingqi Qu, Jing Liu, Wayne Xin Zhao, QiaoQiao She, Hua Wu, Haifeng Wang, Ji-Rong Wen

The Agent’s First Day: Benchmarking Learning, Exploration, and Scheduling in the Workplace Scenarios (2026.findings-acl)

Challenge:	Recent studies show that passage retrieval and passage reranking are important for achieving mutual improvement.
Approach:	They propose a unified listwise training approach for passage retrieval and passage reranking that incorporates a retrieval procedure and a hybrid data augmentation strategy.
Outcome:	The proposed approach improves on both MSMARCO and Natural Questions datasets.

Copied to clipboard

Daocheng Fu, Jianbiao Mei, Rong Wu, Xuemeng Yang, Jia Xu, Ding Wang, Pinlong Cai, Yong Liu, Licheng Wen, Botian Shi

Challenge:	Existing research mainly focuses on performance upper bounds in static environments, overlooking stochastic real-world deployment.
Approach:	They propose a dynamic evaluation environment that simulates a "trainee" agent continuously exploring a novel setting.
Outcome:	The proposed model evaluates agents in a dynamic evaluation environment that simulates a "trainee" agent continuously exploring a novel setting.

TEMP: Taxonomy Expansion with Dynamic Margin Loss through Taxonomy-Paths (2021.emnlp-main)

Copied to clipboard

Zichen Liu, Hongyuan Xu, Yanlong Wen, Ning Jiang, HaiYing Wu, Xiaojie Yuan

MAPO: Boosting Large Language Model Performance with Model-Adaptive Prompt Optimization (2023.findings-emnlp)

Challenge:	Existing taxonomies are unable to maintain coverage due to the rising of new concepts . TEMP uses pre-trained contextual encoders to predict the position of new ideas .
Approach:	They propose a self-supervised taxonomy expansion method that ranks taxonomies by ranking them . they use pre-trained contextual encoders to train the model with dynamic margin loss .
Outcome:	The proposed method outperforms state-of-the-art taxonomy expansion methods by 14.3% and 15.8% on public benchmarks.

Copied to clipboard

Yuyan Chen, Zhihao Wen, Ge Fan, Zhengyu Chen, Wei Wu, Dayiheng Liu, Zhixu Li, Bang Liu, Yanghua Xiao

Kill two birds with one stone: generalized and robust AI-generated text detection via dynamic perturbations (2025.naacl-long)

Challenge:	Existing research emphasizes the importance of adapting prompts to specific tasks, rather than specific LLMs.
Approach:	They propose a model-adaptive prompt optimizer method that optimizes original prompts for each LLM in downstream tasks.
Outcome:	The proposed method can optimize prompts for an LLM in downstream tasks.

Copied to clipboard

Yinghan Zhou, Juan Wen, Wanli Peng, Xue Yiming, ZiWei Zhang, Wu Zhengxian

RAGLAB: A Modular and Research-Oriented Unified Framework for Retrieval-Augmented Generation (2024.emnlp-demo)

Challenge:	Existing methods focus on model generalization or focus on robustness.
Approach:	They propose a model-based AIGT detection method that can be generalized and robust under two adversarial attacks.
Outcome:	The proposed method outperforms state-of-the-art methods for generalization and robustness under two text adversarial attacks.

Copied to clipboard

Xuanwang Zhang, Yun-Ze Song, Yidong Wang, Shuyun Tang, Xinfeng Li, Zhengran Zeng, Zhen Wu, Wei Ye, Wenyuan Xu, Yue Zhang, Xinyu Dai, Shikun Zhang, Qingsong Wen

Digging out Discrimination Information from Generated Samples for Robust Visual Question Answering (2023.findings-acl)

Challenge:	Existing research on Retrieval Augmented Generation (RAG) does not address the problem of hallucinations and real-time updating of knowledge.
Approach:	They propose a modular open-source library to equip LLMs with external knowledge.
Outcome:	The proposed approach reduces the need for expensive open-source tools and lacks fair comparisons between novel RAG algorithms.

Copied to clipboard

Zhiquan Wen, Yaowei Wang, Mingkui Tan, Qingyao Wu, Qi Wu

Implicit Cross-Lingual Rewarding for Efficient Multilingual Preference Alignment (2025.findings-acl)

Challenge:	Existing methods to solve this problem rely on additional annotations and generate negative samples .
Approach:	They propose a method to Dig out Discrimination information from Generated samples to address these limitations.
Outcome:	The proposed method improves on the visual question-answering datasets.

Copied to clipboard

Wen Yang, Junhong Wu, Chen Wang, Chengqing Zong, Jiajun Zhang

Multilinguality

Challenge:	Existing approaches to align English LLMs with human preferences rely on expensive human annotations or advanced multilingual preference alignment models.
Approach:	They propose a method that captures learned preferences from English models by implicit rewards . they annotate preference relations in cross-lingual instruction-following pairs using English .
Outcome:	The proposed approach captures learned preferences from well-aligned English models by implicit rewards and transfers them to other languages through iterative training.

Adaptive Schema-aware Event Extraction with Retrieval-Augmented Generation (2025.findings-emnlp)

Copied to clipboard

Sheng Liang, Hang Lv, Zhihao Wen, Yaxiong Wu, Yongyue Zhang, Hao Wang, Yong Liu

Neuro-Symbolic Query Compiler (2025.findings-acl)

Challenge:	Event extraction is a task in natural language processing that involves identifying and extracting event information from unstructured text.
Approach:	They propose a paradigm that combines schema paraphrasing with schema retrieval-augmented generation.
Outcome:	The proposed paradigm retrieves paraphrased schemas and accurately generates targeted structures.

Copied to clipboard

Yuyao Zhang, Zhicheng Dou, Xiaoxi Li, Jiajie Jin, Yongkang Wu, Zhonghua Li, Ye Qi, Ji-Rong Wen

A Thorough Examination on Zero-shot Dense Retrieval (2023.findings-emnlp)

Challenge:	Retrieval-Augmented Generation (RAG) systems are limited in their ability to process information in open-source environments.
Approach:	They propose a neuro-symbolic framework inspired by linguistic grammar rules and compiler design to formalize complex queries using a minimal yet sufficient Backus-Naur Form grammar.
Outcome:	The proposed framework is based on a backus-naur form grammar and compiler design that maintains completeness while minimizing redundancy.

Copied to clipboard

Ruiyang Ren, Yingqi Qu, Jing Liu, Xin Zhao, Qifei Wu, Yuchen Ding, Hua Wu, Haifeng Wang, Ji-Rong Wen

ATLAS: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex Reasoning (2026.findings-acl)

Challenge:	Recent advances in dense retrieval (DR) models have been shown to be not as competitive as traditional sparse retrieval models in a zero-shot retrieval setting.
Approach:	They propose to examine the zero-shot capability of DR models by analyzing key factors related to source training set and potential bias from target dataset.
Outcome:	The proposed model is not as competitive as sparse retrieval models in a zero-shot retrieval setting.

Copied to clipboard

Jinyang Wu, Guocheng Zhai, Ruihan Jin, Jiahao Yuan, Yuhao Shen, Shuai Zhang, Zhengqi Wen, Jianhua Tao

Inhibitory Attacks on Backdoor-based Fingerprinting for Large Language Models (2026.acl-long)

Challenge:	Existing approaches to optimize large language models with external tools are limited.
Approach:	They propose a dual-path framework for dynamic tool usage in cross-domain complex reasoning . they exploit empirical priors for domain-specific alignment and RL-based multi-step routing .
Outcome:	The proposed framework outperforms closed-source models and existing methods on in-distribution and out-of-distortion tasks.

Copied to clipboard

Hang fu, Wanli Peng, Yinghan Zhou, Jiaxuan Wu, Juan Wen, Xue Yiming

GROVE: A Retrieval-augmented Complex Story Generation Framework with A Forest of Evidence (2023.findings-emnlp)

Challenge:	Backdoor-based LLM fingerprinting is a promising solution for intellectual property protection . however, the vulnerability of existing LLMs for the ensemble scenario is unexplored .
Approach:	They propose two new fingerprinting attack methods to assess the robustness of LLM fingerprinting by token filter attack and sentence verification attack.
Outcome:	The proposed methods inhibit the fingerprint response while maintaining ensemble performance.

Copied to clipboard

Zhihua Wen, Zhiliang Tian, Wei Wu, Yuxin Yang, Yanqi Shi, Zhen Huang, Dongsheng Li

BASES: Large-scale Web Search User Simulation with Large Language Model based Agents (2024.findings-emnlp)

Challenge:	Existing methods for generating stories with complex plots rely on detailed prompts, which inadvertently limit the creative potential of the generated stories.
Approach:	They propose a retrieval-auGmented stoRy generation framework with a fOrest of eVidEnce to enhance stories’ complexity.
Outcome:	The proposed framework enables generating more diverse plotlines from human-written stories.

Copied to clipboard

Ruiyang Ren, Peng Qiu, Yingqi Qu, Jing Liu, Xin Zhao, Hua Wu, Ji-Rong Wen, Haifeng Wang

Implicit Discourse Relation Identification for Open-domain Dialogues (P19-1)

Challenge:	Existing research on web search rely on real-user experiments, which can be costly to scale up.
Approach:	They propose a user simulation framework with LLM-based agents that can generate unique user profiles at scale.
Outcome:	The proposed framework can generate unique user profiles at scale, leading to diverse search behaviors.

Copied to clipboard

Mingyu Derek Ma, Kevin Bowden, Jiaqi Wu, Wen Cui, Marilyn Walker

Dialogue Systems & Conversational Agents

Challenge:	Discourse relation identification is a challenging problem in open-domain dialogue systems . previous work relies on formal text but this data is not suitable for informal dialogue .
Approach:	They propose a method to automatically extract the implicit discourse relation argument pairs from dialogic turns and a pipeline to identify them.
Outcome:	The proposed pipeline extracts argument pairs from dialogic turns and improves it by performing feature ablation and incorporating dialogue features.

Quantification of Large Language Model Distillation (2025.acl-long)

Copied to clipboard

Sunbowen Lee, Junting Zhou, Chang Ao, Kaige Li, Xeron Du, Sirui He, Haihong Wu, Tianci Liu, Jiaheng Liu, Hamid Alinejad-Rokny, Min Yang, Yitao Liang, Zhoufutu Wen, Shiwen Ni

Safety-Utility Conflicts Are Not Global: Surgical Alignment via Head-Level Diagnosis (2026.acl-long)

Challenge:	Existing studies have revealed the robustness degra-dation caused by data distillation.
Approach:	They propose a framework to evaluate and quantify model distillation . they aim to identify identity cognition contradictions and analyse multi-granularity response similarities across models to measure the extent of homogenization.
Outcome:	The proposed framework addresses two key aspects: (1) Identifying identity cognition contradictions to assess discrepancies in how models perceive and represent identity-related information; (2) Analyzing multi-granularity response similarities across models to measure the extent of homogenization.

Copied to clipboard

Wang Cai, Yilin Wen, Jinchang Hou, Du Su, Guoqiu Wang, Zhonghou Lv, Chenfu Bao, Yunfang Wu

Modelling Variability in Human Annotator Simulation (2024.findings-acl)

Challenge:	Existing mitigation strategies rely on global gradient geometry to resolve alignment conflicts . however, they overlook Modular Heterogeneity within Transformers, resulting in suboptimal trade-offs . Conflict-Aware Sparse Tuning (CAST) combines head-level diagnosis with sparse fine-tuning .
Approach:	They propose a framework that integrates head-level diagnosis with sparse fine-tuning to address this limitation.
Outcome:	The proposed framework integrates head-level diagnosis with sparse fine-tuning to reduce alignment conflicts in LLMs.

Copied to clipboard

Wen Wu, Wenlin Chen, Chao Zhang, Phil Woodland

Green & Sustainable NLP

Challenge:	Human annotator simulation (HAS) is a cost-effective alternative to human evaluation tasks.
Approach:	They propose a framework to model human annotation variability via meta-learning . conditional softmax flow model leverages diverse human annotations via meta learning . results demonstrate that method can predict aggregated behaviours of human annotators .
Outcome:	The proposed method achieves state-of-the-art performance on two real-world human evaluation tasks: emotion recognition and toxic speech detection.

SSP: Self-Supervised Post-training for Conversational Search (2023.findings-acl)

Copied to clipboard

Quan Tu, Shen Gao, Xiaolong Wu, Zhao Cao, Ji-Rong Wen, Rui Yan

Dialogue Systems & Conversational Agents

Challenge:	Existing methods to search for information are limited by data scarcity and lack contextual semantic understanding.
Approach:	They propose a new post-training paradigm with three self-supervised tasks to initialize the conversational search model to enhance dialogue structure and contextual semantic understanding.
Outcome:	The proposed method can initialize the conversational search model to enhance dialogue structure and contextual semantic understanding.

Every Document Owns Its Structure: Inductive Text Classification via Graph Neural Networks (2020.acl-main)

Copied to clipboard

Yufeng Zhang, Xueli Yu, Zeyu Cui, Shu Wu, Zhongzhen Wen, Liang Wang