Papers by Zhiwei Li
RFS-Guard: Detecting Reasoning Hallucinations via Cross-Phase Routing Focus in Large Reasoning Models (2026.acl-long)
Copied to clipboard
| Challenge: | Large reasoning models (LRMs) generate intermediate reasoning traces before the final answer, yet they remain vulnerable to reasoning hallucinations such as subtle arithmetic errors. |
| Approach: | They propose a Routing Focus Score (RFS) that measures how strongly cross-step attention routing aligns with semantic proximity derived from hidden-state cosine similarity. |
| Outcome: | The proposed framework detects and localizes hallucinations without external tools or repeated sampling. |
Token Prepending: A Training-Free Approach for Eliciting Better Sentence Embeddings from LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Recent studies have focused on prompt engineering to extract sentence embeddings from large language models (LLMs) but these models are mostly decoder-only and the earlier tokens in the sentence cannot attend to the latter, resulting in biased encoding of sentence information and cascading effects on the final decoded token. |
| Approach: | They propose a plug-and-play and training-free technique that prepends each layer’s decoded sentence embedding to the beginning of the sentence in the next layer’ s input. |
| Outcome: | The proposed technique can significantly improve the performance of existing prompt-based sentence embedding methods across different LLMs while incurring negligible additional inference cost. |
Youling: an AI-assisted Lyrics Creation System (2020.emnlp-demos)
Copied to clipboard
Rongsheng Zhang, Xiaoxi Mao, Le Li, Lin Jiang, Lin Chen, Zhiwei Hu, Yadong Xi, Changjie Fan, Minlie Huang
| Challenge: | Recent studies have focused on a single pass of lyrics generation with little human intervention. |
| Approach: | They propose an AI-assisted lyrics creation system that supports one pass full-text generation and interactive generation modes. |
| Outcome: | The proposed system supports full-text generation and interactive generation modes . it also provides a revision module which enables users to revise undesired lyrics repeatedly. |
Encouraging Good Processes Without the Need for Good Answers: Reinforcement Learning for LLM Agent Planning (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Currently, the dominant end-to-end reinforcement learning paradigm for agents in Large Language Models (LLMs) employs multi-objective optimization that jointly trains both planning and answer summarization capabilities. |
| Approach: | They propose a framework that decouples the training process to enable a focused, single-objective optimization of the planning module. |
| Outcome: | The proposed framework achieves an 8%–12% improvement in planning performance compared to end-to-end baselines. |
Data Mixing Agent: Learning to Re-weight Domains for Continual Pre-training (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for reweighting data mixtures rely on manual designation with certain heuristics based on intuition or empirical results. |
| Approach: | They propose a model-based framework that learns to re-weight domains by reinforcement learning on large quantities of data mixing trajectories with corresponding feedback from an evaluation environment. |
| Outcome: | The proposed framework outperforms baselines in achieving balanced performance across source and target fields and domain spaces without retraining. |
LJPCheck: Functional Tests for Legal Judgment Prediction (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing LJP models fail to evaluate specific aspects of their performance, such as legal fairness and judicial fairness. |
| Approach: | They propose a suite of functional tests for LJP models to comprehend LJp models’ behaviors and offer diagnostic insights. |
| Outcome: | Extensive tests reveal weaknesses in LJP models and provide diagnostic insights. |
EXPLAIN: Enhancing Retrieval-Augmented Generation with Entity Summary (2025.acl-industry)
Copied to clipboard
Yaozhen Liang, Xiao Liu, Jiajun Yu, Zhouhua Fang, Qunsheng Zou, Linghan Zheng, Yong Li, Zhiwei Liu, Haishuai Wang
| Challenge: | Existing document question answering methods reduce inference costs and input tokens. |
| Approach: | They propose a retrieval-augmented generation method that automatically extracts useful entities and generates summaries from documents. |
| Outcome: | The proposed method surpasses baseline retrieval-augmented generation (RAG) and long-context question answering (LC) methods achieve higher accuracy by processing entire documents, but at the cost of increased computational Corresponding authors. |
Multi-Granularity Semantic Revision for Large Language Model Distillation (2026.acl-long)
Copied to clipboard
Xiaoyu Liu, Yun Zhang, Wei Li, Simiao Li, Xudong Huang, Hanting Chen, Yehui Tang, Jie Hu, Zhiwei Xiong, Yunhe Wang
| Challenge: | Existing methods for generating large language models rely on student-generated outputs, which introduce generation errors and misguide the distillation process. |
| Approach: | They propose a multi-granularity semantic revision method for LLM distillation that corrects errors using teacher-generated tokens and re-generates the sequence to minimize errors. |
| Outcome: | The proposed method reduces errors and misguides distillation on student models and improves consistency between teacher and student outputs. |
Mind Reader: Latent User Demand-Guided Content Optimization for Generative Search Engine (2026.acl-long)
Copied to clipboard
Tong Chen, JiaWei Guo, Yuxi Li, Baiming Chen, Houxing Ren, Zhang Zhiwei, Yunxiang Zhang, Hanyang Xia, Kun Liang, Zhaoran Fan
| Challenge: | Generative Search Engines (GSEs) have reshaped information retrieval and Generating Engine Optimization (GEO) emerges to improve the content visibility in GSEs’ responses. |
| Approach: | They propose a method to optimize content to cover latent semantic information of GSEs by decomposing query into diverse perspectives and capturing underlying semantic information. |
| Outcome: | The proposed method outperforms baselines and effectively improves content visibility (with up to 2.44x objective metrics and 1.23x subjective metrics on average). |
DGPO: Beyond Pairwise Preferences with Directional Consistent Groupwise Optimization (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for directional consistency alignment of large language models are limited . a recent study suggests reverse supervision as a complement to forward reasoning . |
| Approach: | They propose a framework that aggregates supervision signals at the group level and explicitly models direction-aware alignment through multi-candidate comparisons. |
| Outcome: | The proposed framework achieves 3.2% accuracy improvement across five benchmarks and multiple datasets. |
Can Large Language Models Mine Interpretable Financial Factors More Effectively? A Neural-Symbolic Factor Mining Agent Model (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing factor mining models are inefficient and inefficient, resulting in a significant challenge to extract interpretable factors. |
| Approach: | They propose a model that integrates the strengths of both neural and symbolic models for factor mining. |
| Outcome: | The proposed model surpasses the SOTA RankIC and RankICIR in predicting S&P 500 returns on real-world stock market data. |
Multi-view Contrastive Learning for Entity Typing over Knowledge Graphs (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to knowledge graph entity typing ignore the way types can be clustered together. |
| Approach: | They propose a method that effectively encodes coarse-grained knowledge from clusters into entity and type embeddings. |
| Outcome: | The proposed method encodes coarse-grained knowledge from clusters into entity and type embeddings. |
History-Aware Hierarchical Transformer for Multi-session Open-domain Dialogue System (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing open-domain dialogue systems conduct one-session conversations, but multi-session MSCs are under-investigated. |
| Approach: | They propose a History-Aware Hierarchical Transformer for multi-session open-domain dialogue . they propose to encode history conversations into a history memory and leverage historical information to generate well-informed responses. |
| Outcome: | The proposed model outperforms baseline models on a large-scale MSC dataset. |
Understanding the Language Model to Solve the Symbolic Multi-Step Reasoning Problem from the Perspective of Buffer Mechanism (2025.findings-emnlp)
Copied to clipboard
Zhiwei Wang, Yunji Wang, Zhongwang Zhang, Zhangchen Zhou, Hui Jin, Tianyang Hu, Jiacheng Sun, Zhenguo Li, Yaoyu Zhang, Zhi-Qin John Xu
| Challenge: | Large language models struggle with complex reasoning tasks, such as mathematical problem-solving. |
| Approach: | They constructed a symbolic multi-step reasoning task to investigate the information propagation mechanisms in Transformer models when solving the task through direct answering and Chain-of-Thought (CoT) reasoning. |
| Outcome: | The proposed algorithm improves on 7 multi-step reasoning datasets, while introducing only 132 trainable parameters. |
TIARA: Multi-grained Retrieval for Robust Question Answering over Large Knowledge Base (2022.emnlp-main)
Copied to clipboard
| Challenge: | KBQA is a challenging area for pre-trained language models due to its extensive space and complexity. |
| Approach: | They propose a model that uses multi-grained retrieval to focus on most relevant KB contexts . constrained decoding is used to control output space and reduce generation errors . |
| Outcome: | The proposed model outperforms existing models on GrailQA and WebQuestionsSP. |
FaStFact: Faster, Stronger Long-Form Factuality Evaluations in LLMs (2025.findings-emnlp)
Copied to clipboard
Yingjia Wan, Haochen Tan, Xiao Zhu, Xinyu Zhou, Zhiwei Li, Qingsong Lv, Changxuan Sun, Jiaqi Zeng, Yi Xu, Jianqiao Lu, Yinhong Liu, Zhijiang Guo
| Challenge: | Prior evaluation pipelines fail to evaluate factuality of long-form LLMs due to inefficiency and costly human assessment. |
| Approach: | They propose a fast and strong evaluation pipeline that can evaluate factuality of long-form LLMs . they propose 'faStFact' to reduce cost of web searching and inference calling . |
| Outcome: | The proposed evaluation pipeline achieves highest alignment with human evaluation and efficiency among existing baselines. |
MultiFinBen: Benchmarking Large Language Models for Multilingual and Multimodal Financial Application (2026.acl-long)
Copied to clipboard
Xueqing Peng, Lingfei Qian, Yan Wang, Ruoyu Xiang, Yueru He, Yang Ren, Mingyang Jiang, Vincent Jim Zhang, Yuqing Guo, Jeff Zhao, Huan He, Yi Han, Yun Feng, Yuechen Jiang, Yupeng Cao, Haohang Li, Yangyang Yu, Xiaoyu Wang, Penglei Gao, Shengyuan Lin, Keyi Wang, Shanshan Yang, Yilun Zhao, Zhiwei Liu, Peng Lu, Jerry Huang, Suyuchen Wang, Triantafillos Papadopoulos, Polydoros Giannouris, Efstathia Soufleri, Nuo Chen, Zhiyang Deng, Heming Fu, Yijia Zhao, Mingquan Lin, Meikang Qiu, Kaleb E Smith, Arman Cohan, Xiao-Yang Liu, Jimin Huang, Guojun Xiong, Alejandro Lopez-Lira, Xi Chen, Junichi Tsujii, Jian-Yun Nie, Sophia Ananiadou, Qianqian Xie
| Challenge: | Existing evaluations of LLMs in finance are text-only, monolingual, and largely saturated by current models. |
| Approach: | They propose a multilingual and multimodal benchmark for evaluating LLMs in real financial contexts. |
| Outcome: | The first expert-annotated multilingual and multimodal benchmark is released . it evaluates 21 leading LLMs and shows they perform better in multilingual settings . |
Transformer-based Entity Typing in Knowledge Graphs (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing knowledge graphs encoding entity types are far from complete, since in real-world applications they are continuously emerging. |
| Approach: | They propose a transformer-based approach to infer plausible entity types by encoding neighbours' information by a local transformer and a global transformer. |
| Outcome: | The proposed approach outperforms the state-of-the-art on two real-world datasets. |
Rethinking Cross-Subject Data Splitting for Brain-to-Text Decoding (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have successfully decoded natural language from non-invasive brain signals . current dataset splitting methods suffer from data leakage problem . |
| Approach: | They propose a right cross-subject data splitting criterion without data leakage for decoding fMRI and EEG signal to text. |
| Outcome: | The proposed method overfits and overestimates brain-to-text decoding models. |
Unraveling the Mystery: Defending Against Jailbreak Attacks Via Unearthing Real Intention (2025.coling-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are increasingly vulnerable to elusive and implicit intentions, causing security risks and compromising user experience. |
| Approach: | They propose a method to detect and mitigate implicit jailbreak attacks using LLMs by unearthing real intentions and a greedy gradient-based algorithm to remove the least important parts of a sentence. |
| Outcome: | The proposed method reduces attacks success rate and Harmful Score while maintaining overall model performance. |
GCRC: A New Challenging MRC Dataset from Gaokao Chinese for Explainable Evaluation (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing machine reading comprehension datasets lack an explainable evaluation of systems' reasoning capabilities. |
| Approach: | They propose a dataset with multi-choice questions that evaluates MRC systems' reasoning process . they use sentence-level relevant supporting facts, error reason of distractors to evaluate MRC . |
| Outcome: | The proposed dataset is more challenging and useful for identifying limitations of existing MRC systems in an explainable way. |
Devil’s Advocate: Anticipatory Reflection for LLM Agents (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Introspection-driven approach equips LLM agents with introspection, enhancing consistency and adaptability in solving complex tasks. |
| Approach: | They propose a zero-shot approach that equips LLM agents with introspection, enhancing consistency and adaptability in solving complex tasks. |
| Outcome: | The proposed approach improves performance and efficiency by reducing the number of trials and plan revisions by 45%. |
R-Judge: Benchmarking Safety Risk Awareness for LLM Agents (2024.findings-emnlp)
Copied to clipboard
Tongxin Yuan, Zhiwei He, Lingzhong Dong, Yiming Wang, Ruijie Zhao, Tian Xia, Lizhen Xu, Binglin Zhou, Fangqi Li, Zhuosheng Zhang, Rui Wang, Gongshen Liu
| Challenge: | Large language models (LLMs) have shown compelling abilities in reasoning, decision-making, and instruction following. |
| Approach: | They propose a benchmark to evaluate the proficiency of large language models (LLMs) in judging and identifying safety risks given agent interaction records. |
| Outcome: | The proposed model outperforms the best-performing model, GPT-4o, while no other models significantly exceed the random. |
Widget Captioning: Generating Natural Language Description for Mobile User Interface Elements (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing tools for examining and fixing missing captions are lacking in mobile UIs. |
| Approach: | They propose a task for automatically generating language descriptions for UI elements from multimodal input including both the image and structural representations of user interfaces. |
| Outcome: | The proposed task can generate captions from image and structural representations of UI elements. |
TEMA: Anchor the Image, Follow the Text for Multi-Modification Composed Image Retrieval (2026.acl-long)
Copied to clipboard
| Challenge: | Composed Image Retrieval (CIR) is an image retrieval paradigm that enables users to retrieve a target image using a multimodal query that consists of a reference image and modification text. |
| Approach: | They propose a text-oriented entity mapping architecture that allows users to use a reference image and modification text to retrieve a target image. |
| Outcome: | The proposed framework is superior in both original and multi-modification scenarios while maintaining an optimal balance between retrieval accuracy and computational efficiency. |
A Zero-Shot Language Agent for Computer Control with Structured Reflection (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Recent works require a model to learn from trace examples of a task via supervised learning or few/many-shot prompting. |
| Approach: | They propose a model that iteratively learns from its mistakes via self-reflection and structured thought management. |
| Outcome: | The proposed model outperforms previous models on easy tasks with more efficient reasoning and self-reflection. |
Towards Scalable Lightweight GUI Agents via Multi-role Orchestration (2026.findings-acl)
Copied to clipboard
Ziwei Wang, Junjie Zheng, Leyang Yang, Sheng Zhou, Xiaoxuan Tang, Fang Zhouhua, Zhiwei Liu, Dajun Chen, Yong Li, Jiajun Bu
| Challenge: | Advanced GUI agents suffer from prohibitive deployment costs on resource-constrained devices. |
| Approach: | They propose a lightweight GUI agent with GUI-specific knowledge and task scalability . LAMO-3B supports monolithic execution and MAS-style orchestration . |
| Outcome: | The proposed GUI agent LAMO-3B supports monolithic execution and MAS-style orchestration. |
Abstract, Rationale, Stance: A Joint Model for Scientific Claim Verification (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing scientific claim verification models have problems of error propagation among modules and lack of sharing valuable information among modules. |
| Approach: | They propose an approach that jointly learns the modules for the three tasks with a machine reading comprehension framework by including claim information. |
| Outcome: | The proposed approach outperforms existing models on the SciFact dataset on the three tasks of abstract retrieval, rationale selection and stance prediction. |
CONSTRUCTURE: Benchmarking CONcept STRUCTUre REasoning for Multimodal Large Language Models (2024.findings-emnlp)
Copied to clipboard
Zhiwei Zha, Xiangru Zhu, Yuanyi Xu, Chenghua Huang, Jingping Liu, Zhixu Li, Xuwu Wang, Yanghua Xiao, Bei Yang, Xiaoxiao Xu
| Challenge: | Existing multimodal large language models lack the ability to perceive the visual world with a deep concept structure cognition. |
| Approach: | They propose a concept-level benchmark to assess MLLMs’ hierarchical concept understanding and reasoning abilities. |
| Outcome: | The proposed model outperforms state-of-the-art models in concept structure reasoning evaluation. |