Papers by Zhunchen Luo
IS-CoT: Breaking the Long-form Generation Collapse via Interleaved Structural Thinking (2026.acl-long)
Copied to clipboard
Zechen Sun, Yuyang Sun, Zecheng Tang, Juntao Li, Wenpeng Hu, Wenliang Chen, Zhunchen Luo, Guotong Geng, Min Zhang
| Challenge: | Existing models with reasoning capabilities suffer from a severe length collapse in open-ended writing . |
| Approach: | They propose a framework that embeds a dynamic plan-write-reflect cycle into the generation process and train a model with interleaved reasoning traces. |
| Outcome: | The proposed framework achieves state-of-the-art performance on long-form benchmarks compared to other models on the same dataset. |
DisCal: Distribution-Aware Calibration for Mathematical Reasoning Under Character-Level Noisy Inputs (2026.acl-long)
Copied to clipboard
Bo Zhang, Jiawei Zhang, Cong Gao, Bingxu Han, Minghao Hu, Jun Zhang, Yunbo Cao, Zhunchen Luo, Wen Yao, Guotong Geng, Zhong Wang
| Challenge: | Existing methods for calibration of large reasoning models (LRMs) focus on clean inputs, leaving noise unexplored. |
| Approach: | They propose a confidence calibration framework for character-level noisy inputs that extracts uncertainty signals from both the empirical answer distribution and the model’s predictive distribution and integrates them via a learned calibrator. |
| Outcome: | Experiments on multiple mathematical reasoning benchmarks show that DisCal outperforms existing calibration methods under noisy inputs, reducing expected calibration error (ECE) by up to 39.21% and improving Area Under the Receiver Operating Characteristic Curve (AUROC) by 31.44%. |
A Context-based Framework for Modeling the Role and Function of On-line Resource Citations in Scientific Literature (D19-1)
Copied to clipboard
| Challenge: | Existing academic search engines cannot detect relevant papers where a resource is mentioned. |
| Approach: | They propose a framework to model the role and function of on-line resource citations . they construct a dataset SciRes, which includes 3,088 manually annotated resource contexts based on a multi-task framework . |
| Outcome: | The proposed model achieves the best results on both the classification task and recommendation task. |
Characterizing and Verifying Scientific Claims: Qualitative Causal Structure is All You Need (2023.emnlp-main)
Copied to clipboard
| Challenge: | a scientific claim verification requires thorough examination and assessment to ascertain its validity . attention architectures and pre-trained language models fail to establish a comprehensive chain of causal inference . |
| Approach: | They propose a qualitative causal structure-based graph neural network model to facilitate causal reasoning across relevant causally-potent factors. |
| Outcome: | The proposed model outperforms state-of-the-art models by incorporating semantic features . the proposed model is based on a qualitative causal structure . |
Identifying Principals and Accessories in a Complex Case based on the Comprehension of Fact Description (2020.acl-main)
Copied to clipboard
| Challenge: | Existing studies on complex criminal cases with multiple defendants only focus on the simple cases with one defendant. |
| Approach: | They propose to model the defendants with behavioral semantic information and statistical characteristics, then learning the importances of defendants within a learning-to-rank framework. |
| Outcome: | The proposed model can model the defendants’ impacts in a complex case. |
Unveiling the Potential of BERT-family: A New Recipe for Building Scalable, General and Competitive Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Generative large language models (LLMs) have significantly influenced various aspects of society, reshaping how we access and interact with information and knowledge. |
| Approach: | They propose a pre-training task that helps BERT-family excel in wider applications . they also explore the integration of cutting-edge technologies into their models to further enhance their capabilities. |
| Outcome: | The proposed model exhibits performance levels comparable to current SOTA LLMs across a spectrum of tasks. |
When Rules Learn: A Self-Evolving Agent for Legal Case Retrieval (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing dense retrieval methods have achieved notable progress, but their effectiveness in legal case retrieval remains limited. |
| Approach: | They propose a self-evolving framework for rule-driven query rewriting that enhances BM25 without any parameter training. |
| Outcome: | The proposed framework outperforms non-evolutionary baselines, including human-designed rules and greedy rule selection, especially when powered by a high-capacity core LLM. |
Uncovering Argumentative Flow: A Question-Focus Discourse Structuring Framework (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing structure modeling approaches fail to capture the author’s rhetorical intent and reasoning process. |
| Approach: | They propose a Question-Focus discourse structuring framework that explicitly models the underlying argumentative flow by anchoring each argumentative unit to a guiding question and a set of attentional foci. |
| Outcome: | The proposed framework outperforms baseline models and curated models on an argument reconstruction task in Chinese think-tank articles and claims coverage. |
TP-Detector: Detecting Turning Points in the Engineering Process of Large-scale Projects (2023.emnlp-demo)
Copied to clipboard
| Challenge: | Extensive experiments demonstrate the effectiveness of our proposed method on a constructed dataset compared to baseline methods. |
| Approach: | They propose a novel task of detecting turning points in the engineering process of large-scale projects by treating news streams as a window with multiple instances. |
| Outcome: | The proposed mode is able to detect transitions in news streams with multiple instances. |
Jointly Multiple Events Extraction via Attention-based Graph Information Aggregation (D18-1)
Copied to clipboard
| Challenge: | Event extraction is of practical utility in natural language processing . it is common that multiple events exist in the same sentence, causing difficulties in extracting them . |
| Approach: | They propose a framework to jointly extract multiple event triggers and arguments . they introduce syntactic shortcut arcs to enhance information flow and attention-based graph convolution networks to model graph information. |
| Outcome: | The proposed framework achieves competitive results compared with state-of-the-art methods. |
Interpretable Charge Predictions for Criminal Cases: Learning to Generate Court Views from Fact Descriptions (N18-1)
Copied to clipboard
| Challenge: | Existing work on court view generation from fact descriptions has improved the working efficiency of legal assistant systems. |
| Approach: | They propose to decode court views conditioned on encoded charge labels from the fact description in a criminal case to improve interpretability of charge prediction systems. |
| Outcome: | The proposed model can generate court views conditioned on encoded charge labels. |
Real-time Scholarly Retweeting Prediction System (C18-2)
Copied to clipboard
| Challenge: | a scholarly retweeting prediction system is proposed to predict scholarly tweets . re-tweening is an action of reposting others' tweet by using the reretwet button on Twitter . |
| Approach: | They propose a real-time scholarly retweeting prediction system that retrieves scholarly tweets which will be re-tweeled. |
| Outcome: | The proposed system outperforms baseline systems and can predict scientific impact in real-time. |
SafeConf: A Confidence-Calibrated Safety Self-Evaluation Method for Large Language Models (2025.findings-emnlp)
Copied to clipboard
Bo Zhang, Cong Gao, Linkang Yang, Bingxu Han, Minghao Hu, Zhunchen Luo, Guotong Geng, Xiaoying Bai, Jun Zhang, Wen Yao, Zhong Wang
| Challenge: | Large language models (LLMs) have many advantages but they also pose significant safety risks. |
| Approach: | They propose a method to enhance the safety self-evaluation capability of LLMs . they perform semantic mutations on the original safety evaluation questions . |
| Outcome: | The proposed method improves safety self-evaluation accuracy by 5.86% and 7.79% over baseline methods on Chinese and English datasets. |
Interpretable Rationale Augmented Charge Prediction System (C18-2)
Copied to clipboard
| Challenge: | Existing studies treat charge prediction as a text classification problem, but in the field of justice, every decision may be a matter of life and death. |
| Approach: | They propose to extract readable rationales from text and then create a rationale augmented classification model to enhance the prediction accuracy. |
| Outcome: | The proposed system can extract readable rationales in a high consistency with manual annotation and is comparable with the attention model in prediction accuracy. |
Dynamic Evil Score-Guided Decoding: An Efficient Decoding Framework For Red-Team Model (2025.findings-acl)
Copied to clipboard
Cong Gao, Bo Zhang, Linkang Yang, Minghao Hu, Zhunchen Luo, Xiaoying Bai, Guotong Geng, Jun Zhang, Yunhua Xue
| Challenge: | Existing red-teaming methods require expensive fine-tuning, especially for large LLMs. |
| Approach: | They propose a red-teaming method that uses an ‘evil score’ to evaluate the potential of tokens to contribute to harmful outputs during decoding. |
| Outcome: | The proposed method achieves an ASR of 92.83% on the Llama-3.2-3B-Instruct model, compared to 83.48% with adversarial fine-tuning while using less computational resources. |
CRST: a Claim Retrieval System in Twitter (C18-2)
Copied to clipboard
| Challenge: | CRST retrieves tweets containing arguments for controversial topics from Twitter. |
| Approach: | They propose a system that retrieves tweets containing claims for a given topic from Twitter. |
| Outcome: | The proposed system outperforms existing claims retrieval and argument mining systems. |
Improved Training of Deep Text Clustering (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for deep clustering optimization with shallow models have limited performance due to poor power of feature learning. |
| Approach: | They propose a general deep clustering optimization method that leverages information feedback to construct generalized labels to optimize the deep model. |
| Outcome: | The proposed method reduces the impact of noise on the clustering process by using correlation relationship between the samples. |