Papers by Yaxin Li
Holistic Automated Red Teaming for Large Language Models through Top-Down Test Case Generation and Multi-turn Interaction (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches focus on improving attack success rates while overlooking the need for comprehensive test case coverage. |
| Approach: | They propose a top-down approach to automated red teaming that scales up the diversity of test cases using an extensible, fine-grained risk taxonomy. |
| Outcome: | The proposed approach scales up the diversity of test cases using a top-down approach based on an extensible, fine-grained risk taxonomy and leverages reinforcement learning techniques to facilitate multi-turn adversarial probing in a human-like manner. |
DraDDP: A Multimodal Multi-Party Dialogue Discourse Parsing Dataset (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on multi-party dialogue discourse parsing focus on textual modality and two-party dialog . et al., 2016) focused on text-based discourse parses, ignoring the complexity and richness of multimodal interactions in real-world scenarios. |
| Approach: | They construct the first publicly available English multimodal dataset for multi-party dialogue discourse parsing based on American TV dramas. |
| Outcome: | The proposed dataset contains 495 dialogue segments with 6,374 utterances and 9.1 hours of parallel video content, covering rich multi-party interaction scenarios. |
Legal Mathematical Reasoning with LLMs: Procedural Alignment through Two-Stage Reinforcement Learning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing legal mathematical reasoning models lack structured numerical reasoning . existing models perform poorly on LexNum, while LexPam improves both mathematical accuracy and legal coherence. |
| Approach: | They propose a legal mathematical reasoning benchmark LexNum and LexPam to address this problem . LexPam is a two-stage reinforcement learning framework for efficient legal reasoning training. |
| Outcome: | The proposed framework improves mathematical accuracy and legal coherence . it also improves legal cohesion and generalizes effectively across tasks and domains. |
QAP: A Quantum-Inspired Adaptive-Priority-Learning Model for Multimodal Emotion Recognition (2023.findings-acl)
Copied to clipboard
| Challenge: | Experimental results show that multimodal emotion recognition is a state-of-the-art technique . textual, visual and acoustic modalities are involved in multimodal video emotion recognition . |
| Approach: | They propose a quantum-inspired adaptive-priority-learning model to address the challenges . they use quantum state to model modal features and Q-attention to integrate three modalities . |
| Outcome: | Experimental results show that QAP improves on previous models. |
Improving Dialogue Discourse Parsing through Discourse-aware Utterance Clarification (2025.acl-long)
Copied to clipboard
| Challenge: | Extensive experiments on the STAC and Molweni datasets demonstrate that our approach effectively resolves ambiguities and significantly outperforms the state-of-the-art (SOTA) baselines. |
| Approach: | They propose a Discourse-aware Clarification Module (DCM) that generates clarifications for the parser through systematic clarification type reasoning and discourse goal reasoning. |
| Outcome: | Extensive experiments on the STAC and Molweni datasets demonstrate that the proposed module significantly outperforms the state-of-the-art (SOTA) framework. |
Enhancing Goal-oriented Proactive Dialogue Systems via Consistency Reflection and Correction (2025.acl-long)
Copied to clipboard
| Challenge: | Unlike traditional dialogue systems, goal-oriented proactive dialogue systems focus on achieving specific objectives by actively guiding and anticipating user needs. |
| Approach: | They propose a model-agnostic two-stage Consistency Reflection and Correction framework that allows the model to reflect on discrepancies between generated responses and dialogue contexts and suggest possible corrections. |
| Outcome: | The proposed framework significantly improves the consistency between generated responses and dialogue contexts on three datasets. |
A Distance-Aware Multi-Task Framework for Conversational Discourse Parsing (2022.coling-1)
Copied to clipboard
| Challenge: | Existing studies have focused on graph-based and transition-based discourse parsing, but no study has investigated the advantages of both paradigms for conversational discourse paring. |
| Approach: | They propose a distance-aware multi-task framework that incorporates the strengths of transition-based paradigms to facilitate conversational discourse parsing. |
| Outcome: | The proposed framework improves the graph-based paradigm on long-distance dependency links. |
Incomplete Utterance Rewriting with Editing Operation Guidance and Utterance Augmentation (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing generation methods on Incomplete Utterance Rewriting (IUR) can generate coherent utterances, but they often include irrelevant and redundant tokens in rewritten utteras . |
| Approach: | They propose a multi-task learning framework that uses editing operation labels to guide generation model to focus on critical tokens in dialogue context. |
| Outcome: | The proposed model outperforms state-of-the-art models on open-domain and task-oriented dialogues on three datasets. |
Improving Multi-party Dialogue Generation via Topic and Rhetorical Coherence (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies on multi-party dialogue generation focus on the reply-to structure of dialogue histories, but they neglect the coherence between generated responses and target utterances. |
| Approach: | They propose a Reinforcement Learning approach emphasizing Topic and Rhetorical Coherence to enhance the model's perception of coherence with the target utterance. |
| Outcome: | The proposed approach significantly outperforms the state-of-the-art baselines on two popular datasets. |
TCPO: Thought-Centric Preference Optimization for Effective Embodied Decision-making (2025.emnlp-main)
Copied to clipboard
Kechen Jiao, Zhirui Fang, Jiahao Liu, Bei Li, Qifan Wang, Xinyu Liu, Junhao Ruan, Zhongjian Qiao, Yifan Zhu, Yaxin Xu, Jingang Wang, Xiu Li
| Challenge: | Existing post-SFT methods for embodied AI are constrained by sparse rewards and action-only optimization, resulting in low sample efficiency, poor consistency, and model degradation. |
| Approach: | They propose to integrate Thought-Centric Preference Optimization (TCPO) into embodied decision-making by transforming sparse reward signals into richer step sample pairs. |
| Outcome: | The proposed approach achieves an average success rate of 26.67% in the ALFWorld environment, and a 6% improvement over RL4VLM. |
Faithful Inference Chains Extraction for Fact Verification over Multi-view Heterogeneous Graph with Causal Intervention (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods for fact verification do not extract faithful inference chains due to the diversity of relation paths. |
| Approach: | They propose a multi-view heterogeneous Graph with causal intervention to extract evidence graphs from the knowledge graph. |
| Outcome: | The proposed model provides precise evidence graphs and achieves state-of-the-art performance on the public KG-based fact verification dataset FactKG. |
LLMSurgeon: Diagnosing Data Mixture of Large Language Models (2026.acl-long)
Copied to clipboard
Yaxin Luo, Jiacheng Cui, Xiaohan Zhao, Xinyi Shang, Jiacheng Liu, Xinyue Bi, Zhaoyi Li, Zhiqiang Shen
| Challenge: | a lack of transparency in large language models makes auditing their "digital DNA" difficult. |
| Approach: | They propose a framework that casts DMS as an inverse problem under label-shift assumption . they propose LLMScan, a recipe-verifiable evaluation suite built from open-source LLMs . |
| Outcome: | The proposed framework casts DMS as an inverse problem under label-shift assumption . compared with existing frameworks, it recovers domain mixtures with high fidelity . |
Exploring Memorization in Fine-tuned Language Models (2024.acl-long)
Copied to clipboard
Shenglai Zeng, Yaxin Li, Jie Ren, Yiding Liu, Han Xu, Pengfei He, Yue Xing, Shuaiqiang Wang, Jiliang Tang, Dawei Yin
| Challenge: | Existing studies have shown that pre-trained langauge models tend to memorize and regenerate segments of their pre-training corpus when prompted appropriately. |
| Approach: | They conduct the first comprehensive analysis to explore language models’ memorization during fine-tuning across tasks. |
| Outcome: | The proposed analysis shows that memorization presents a strong disparity among different fine-tuning tasks. |
Enhancing Goal-oriented Proactive Dialogue Systems via Dynamic Multi-dimensional Consistency Optimization (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing work on goal-oriented proactive dialogue systems failed to address the multi-dimensional consistency issue between generated responses and key contextual elements. |
| Approach: | They propose a Dynamic Multi-dimensional Consistency Reinforcement Learning framework which measures the impact of each consistency dimension on overall dialogue quality and provides feedback to improve response quality. |
| Outcome: | The proposed framework significantly improves the consistency of generated responses on two datasets. |
“Yes, My LoRD.” Guiding Language Model Extraction with Locality Reinforced Distillation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for model extraction attacks on large language models are inadequate . existing methods neglect the inconsistency between training tasks and LLM alignment . |
| Approach: | They propose a model extraction algorithm that uses a policy-gradient-style training task to guide the crafting of preference for the local model. |
| Outcome: | The proposed algorithm reduces query complexity while mitigating watermark protection . it can extract various state-of-the-art commercial LLMs while minimizing query complexity . |
Simulating Dual-Process Thinking in Dialogue Topic Shift Detection (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods for topic shift detection focus on shallow local reasoning, overlooking the importance of considering the global historical structure and local details to elucidate the underlying causes of topic shift. |
| Approach: | They propose a dual-process theory for dialogue topic shift detection that employs Large Language Models to extract and store the global topic structure of historical dialogue, while a reasoning module introduces a LLM to generate reasoning samples between the response and the most recent topic of historical dialog. |
| Outcome: | The proposed framework outperforms the state-of-the-art on three public datasets and is based on a dual-process theory. |
TrojanSQL: SQL Injection against Natural Language Interface to Database (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies on text-to-SQL systems have not investigated its security aspects . however, how to implement such attacks remains an open question. |
| Approach: | They propose a backdoor-based SQL injection framework for text-to-SQL systems that uses boolean-based injection and union-based injecting techniques to exploit SQL injection vulnerabilities. |
| Outcome: | The proposed framework can produce harmful SQL statements invalidating user queries or compromise sensitive information about the database. |
Uncertainty-Aware Iterative Preference Optimization for Enhanced LLM Reasoning (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for enhancing the performance of large language models require expensive manual annotations. |
| Approach: | They propose an offline direct preference optimization method that collects preference pairs through iterative sampling and execution feedback to improve model confidence. |
| Outcome: | The proposed method improves performance on three reasoning tasks and shows a 3.6% improvement over the standard method. |
Document-level Relationship Extraction by Bidirectional Constraints of Beta Rules (2023.emnlp-main)
Copied to clipboard
| Challenge: | Document-level Relation Extraction (DocRE) aims to extract relations among entity pairs in documents. |
| Approach: | They propose a logic constraint framework that uses bidirectional constraints to model rules by beta contribtion and reconstruct rule consistency loss by bidirectional constraint. |
| Outcome: | The proposed framework outperforms existing models in relation extraction performance and logical consistency. |
Improving Dialogue Discourse Parsing via Reply-to Structures of Addressee Recognition (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to learn dialogue discourse parsing with related tasks require additional annotation, thus limiting their generality. |
| Approach: | They propose a multitasking framework that integrates dialogue discourse parsing with addressee recognition to reflect relation-based structure of dialogue. |
| Outcome: | The proposed framework outperforms baselines on the Molweni and STAC datasets. |
Enhancing Multi-party Dialogue Discourse Parsing with Explanation Generation (2025.coling-main)
Copied to clipboard
| Challenge: | Multi-party dialogue discourse parsing is an important and challenging task in natural language processing. |
| Approach: | They propose a model to integrate external knowledge from Large Language Models to analyze dialogue discourse structures and semantic relations between utterances in multi-party conversations. |
| Outcome: | The proposed model outperforms the state-of-the-art (SOTA) models on two public datasets. |
Mitigating Shortcut Learning via Smart Data Augmentation based on Large Language Model (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods to improve shortcut learning performance are limited by manual definition of shortcuts and inherent confirmation bias during model training. |
| Approach: | They propose a method of Smart Data Augmentation based on Large Language Models to identify shortcuts and generate their anti-shortcut counterparts. |
| Outcome: | The proposed method shows an improvement of 5.61% across various natural language processing tasks. |
Uncovering the Potential of ChatGPT for Discourse Analysis in Dialogue: An Empirical Study (2024.lrec-main)
Copied to clipboard
| Challenge: | Large language models have shown remarkable capability in many downstream tasks, yet their ability to understand discourse structures of dialogues remains less explored. |
| Approach: | They aim to systematically inspect ChatGPT’s performance in two discourse analysis tasks: topic segmentation and discourse parsing. |
| Outcome: | The proposed model can give more reasonable topic structures than human annotations but only linearly parses the hierarchical rhetorical structures. |
Two-stage Incomplete Utterance Rewriting on Editing Operation (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods to generate rewritten utterances based on dialogue context ignore coreference and ellipsis in dialogues. |
| Approach: | They propose a framework where the first stage generates editing operations and the second stage rewrites incomplete utterances utilizing the generated editing operations. |
| Outcome: | The proposed framework outperforms the existing models on three IUR datasets. |
Not Just Classification: Recognizing Implicit Discourse Relation on Joint Modeling of Classification and Generation (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods of implicit discourse relation recognition (IDRR) focus on three aspects: enhancing discourse units representation, enhancing semantic interaction, and joint learning with other tasks. |
| Approach: | They propose a joint model to recognize the relation label and generate the target sentence containing the meaning of relations simultaneously. |
| Outcome: | The proposed model achieves the best performance against several state-of-the-art systems on Chinese and English datasets. |
AMOA: Global Acoustic Feature Enhanced Modal-Order-Aware Network for Multimodal Sentiment Analysis (2022.coling-1)
Copied to clipboard
| Challenge: | Existing methods treat three modal features equally, without distinguishing the importance of different modalities. Existing models split the video into frames, leading to missing the global acoustic information. |
| Approach: | They propose a global Acoustic feature enhanced Modal-Order-Aware network to address these problems. |
| Outcome: | The proposed model outperforms state-of-the-art models on two public datasets. |
Non-Emotion-Centric Empathetic Dialogue Generation (2025.coling-main)
Copied to clipboard
| Challenge: | Empathy is a social psychology theory that enables individuals to comprehend each other's experiences and emotions, thereby fostering more intimate interpersonal relationships. |
| Approach: | They propose a framework for empathetic dialogue generation based on contrastive learning and context-sensitive entity and social commonsense that punishes responses with incorrect emotions and improves the quality of emotions. |
| Outcome: | The proposed framework improves the quality of empathetic generation and generates more diverse responses in comparison with the state-of-the-art baselines. |