Papers by Zhiyu Zhang
TwiUSD: A Benchmark Dataset and Structure-Aware LLM Framework for User Stance Detection (2026.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks for political user-level stance detection rely on noisy heuristics or distant supervision. |
| Approach: | They propose a large-scale, expert-annotated benchmark for political user-level stance detection with explicit social network structure that integrates user content and followee signals. |
| Outcome: | The proposed framework outperforms baselines in terms of quality and reliability. |
CodeFlowBench: A Multi-turn, Iterative Benchmark for Complex Code Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Modern software development demands code that is maintainable, testable, and scalable by organizing the implementation into modular components with iterative reuse of existing codes. |
| Approach: | They propose a benchmark to evaluate LLMs' ability to perform codeflow by reusing existing functions over multiple turns. |
| Outcome: | The proposed benchmarks show that LLMs perform significantly worse in multi-turn codeflow scenarios and that their performance inversely correlates with dependency complexity. |
Preference Learning Unlocks LLMs’ Psycho-Counseling Skills (2026.findings-acl)
Copied to clipboard
| Challenge: | Current LLMs struggle to consistently provide effective responses to client speeches due to the lack of supervision from high-quality real psycho-counseling data. |
| Approach: | They propose to use a dataset to evaluate therapists' responses to client speeches using a set of professional and comprehensive principles to evaluate their responses. |
| Outcome: | The proposed model achieves an impressive win rate of 87% against GPT-4o. |
REIC: RAG-Enhanced Intent Classification at Scale (2025.emnlp-industry)
Copied to clipboard
Ziji Zhang, Michael Yang, Zhiyu Chen, Yingying Zhuang, Shu-Ting Pi, Qun Liu, Rajashekar Maragoud, Vy Nguyen, Anurag Beniwal
| Challenge: | Accurate intent classification is critical for efficient routing in customer service . however, as companies expand their product lines, intent classification faces scalability challenges . |
| Approach: | They propose a retrieval-augmented generation Enhanced Intent Classification approach which leverages retrieval augmented generation to integrate relevant knowledge into a model. |
| Outcome: | The proposed approach outperforms fine-tuning, zero-shot, and few-shot methods on real-world datasets. |
KPatch: Knowledge Patch to Pre-trained Language Model for Zero-Shot Stance Detection on Social Media (2024.lrec-main)
Copied to clipboard
Shuohao Lin, Wei Chen, Yunpeng Gao, Zhishu Jiang, Mengqi Liao, Zhiyu Zhang, Shuyuan Zhao, Huaiyu Wan
| Challenge: | Existing knowledge injection methods fail to understand the semantics of tweets . |
| Approach: | They propose a method to flexibly inject knowledge into a pre-trained language model and adaptively expand tweets context. |
| Outcome: | The proposed method is based on two training stages to flexibly inject knowledge into the pre-trained language model and adaptively expand tweets context. |
CBT-Bench: Evaluating Large Language Models on Assisting Cognitive Behavior Therapy (2025.naacl-long)
Copied to clipboard
Mian Zhang, Xianjun Yang, Xinlu Zhang, Travis Labrum, Jamie C. Chiu, Shaun M. Eack, Fei Fang, William Yang Wang, Zhiyu Chen
| Challenge: | Existing research has explored mental health condition classifications, empathetic conversations, and chatbots designed for simple discourse structures. |
| Approach: | They propose a benchmark for systematic evaluation of cognitive behavioral therapy assistance using Large Language Models (LLMs). |
| Outcome: | The proposed benchmark includes three levels of tasks covering key aspects of cognitive behavioral therapy that could be enhanced through AI assistance. |
MT-Video-Bench: A Holistic Video Understanding Benchmark for Evaluating Multimodal LLMs in Multi-Turn Dialogues (2026.findings-acl)
Copied to clipboard
Yaning Pan, Qianqian Xie, Guohui Zhang, Zekun Moore Wang, Yongqian Wen, Yuanxing Zhang, Haoxuan Hu, Zhiyu Pan, Yibing Huang, Zhidong Gan, Yonghong Lin, An Ping, Shihao Li, Yanghai Wang, Tianhao Peng, Jiaheng Liu
| Challenge: | Existing evaluation benchmarks for Multimodal Large Language Models (MLLMs) focus on single-turn question answering, overlooking the complexity of multi-turn dialogues in real-world scenarios. |
| Approach: | They propose a video understanding benchmark for MLLMs in multi-turn dialogues that assesses six core competencies that focus on perceptivity and interactivity. |
| Outcome: | The MT-Video-Bench evaluates 1,000 multi-turn dialogues from diverse domains and reveals significant performance discrepancies and limitations in handling multi-turned video dialogues. |
StreamMeCo: Long-Term Agent Memory Compression for Efficient Streaming Video Understanding (2026.findings-acl)
Copied to clipboard
Junxi Wang, Te Sun, Jiayi Zhu, Junxian Li, Haowen Xu, Zichen Wen, Xuming Hu, Zhiyu li, Linfeng Zhang
| Challenge: | StreamMeCo is an efficient Stream Agent Memory Compression framework for video understanding. |
| Approach: | They propose an efficient Stream Agent Memory Compression framework that evicts redundant memory nodes and introduces a time-decay memory retrieval mechanism to mitigate performance degradation. |
| Outcome: | The proposed framework achieves 1.87 speedup in memory retrieval while delivering an average accuracy improvement of 1.0% on three challenging benchmark datasets. |
ObfusLM: Privacy-preserving Language Model Service against Embedding Inversion Attacks (2025.acl-long)
Copied to clipboard
Yu Lin, Ruining Yang, Yunlong Mao, Qizhi Zhang, Jue Hong, Quanwei Cai, Ye Wu, Huiqi Liu, Zhiyu Chen, Bing Duan, Sheng Zhong
| Challenge: | Recent studies show that obfuscation techniques for MLaaS are susceptible to embedding inversion attacks (EIAs). |
| Approach: | They propose a model obfuscation framework that protects client inputs from embedding inversion attacks by obliviously obbing models. |
| Outcome: | The proposed framework outperforms existing works in utility by 10% with a nearly 80% resistance rate against embedding inversion attacks. |
Is Grokking Worthwhile? Functional Analysis and Transferability of Generalization Circuits in Transformers (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models struggle with the "curse of two-hop reasoning" in compositional tasks. |
| Approach: | They propose to form a "Generalization Circuit" during a prolonged "grokking" phase . they argue that grokkking is the process of integrating memorized atomic facts into an easy-acquire reasoning path. |
| Outcome: | The proposed model is superior to non-grokked models, but it requires a large computational cost . the study shows that grokking is not the sudden acquisition of a new reasoning paradigm . |
PHMOSpell: Phonological and Morphological Knowledge Guided Chinese Spelling Check (2021.acl-long)
Copied to clipboard
| Challenge: | False gram and phonological errors make Chinese spelling check difficult . a novel end-to-end trainable model outperforms existing methods . |
| Approach: | They propose a trainable Chinese spelling check model that integrates phonological and visual information into a pre-trained language model. |
| Outcome: | The proposed model outperforms existing state-of-the-art models on three benchmarks. |
UBench: Benchmarking Uncertainty in Large Language Models with Multiple Choice Questions (2025.findings-acl)
Copied to clipboard
Xunzhi Wang, Zhuowei Zhang, Gaonan Chen, Qiongyu Li, Bitong Luo, Zhixin Han, Haotian Wang, Zhiyu Li, Hang Gao, Mengting Hu
| Challenge: | Existing methods for benchmarking the uncertainty of large language models face challenges . existing methods require internal model access, additional training, or high computational costs . |
| Approach: | They propose a new benchmark for evaluating the uncertainty of large language models based on confidence intervals . UBench encompasses 11,978 multiple choice questions spanning knowledge, language, understanding, and reasoning capabilities. |
| Outcome: | The proposed method outperforms existing methods for benchmarking the uncertainty of large language models. |
Adversarial Preference Learning for Robust LLM Alignment (2025.findings-acl)
Copied to clipboard
Yuanfu Wang, Pengyu Wang, Chenyang Xi, Bo Tang, Junyi Zhu, Wenqiang Wei, Chen Chen, Chao Yang, Jingfeng Zhang, Chaochao Lu, Yijun Niu, Keming Mao, Zhiyu Li, Feiyu Xiong, Jie Hu, Mingchuan Yang
| Challenge: | Modern language models rely on Reinforcement Learning from Human Feedback (RLHF) to encourage safe behaviors, but they remain vulnerable to adversarial attacks due to three key limitations: (1) the inefficiency and high cost of human annotation; (2) the vast diversity of potential adversarials; and (3) the risk of feedback bias and reward hacking. |
| Approach: | They propose an iterative adversarial training method that incorporates three key innovations to address these challenges. |
| Outcome: | Experiments on Mistral-7B-Instruct-v0.3 show that the proposed method significantly enhances robustness and reduces harmful outputs from 5.88% to 0.43%. |
A Generative Adaptive Replay Continual Learning Model for Temporal Knowledge Graph Reasoning (2025.acl-long)
Copied to clipboard
| Challenge: | Existing Continual Learning (CL)-based Temporal Knowledge Graph Reasoning methods are incomplete and reorganize historical facts without preserving historical knowledge. |
| Approach: | They propose a method which generates and adaptively replays historical entity distributions from the whole historical context. |
| Outcome: | The proposed method outperforms baselines in reasoning and mitigating forgetting. |
TextEE: Benchmark, Reevaluation, Reflections, and Future Challenges in Event Extraction (2024.findings-acl)
Copied to clipboard
Kuan-Hao Huang, I-Hung Hsu, Tanmay Parekh, Zhiyu Xie, Zixuan Zhang, Prem Natarajan, Kai-Wei Chang, Nanyun Peng, Heng Ji
| Challenge: | Recent studies suggest that event extraction evaluations may not accurately reflect the true performance. |
| Approach: | They propose a standardized, fair, and reproducible benchmark for event extraction . they use standardized scripts and splits for 16 datasets spanning eight domains . |
| Outcome: | The proposed benchmarks show that they struggle to achieve satisfactory performance. |
CloneMem: Benchmarking Long-Term Memory for AI Clones (2026.acl-long)
Copied to clipboard
| Challenge: | Existing memory benchmarks rely on user–agent conversational histories, which are temporally fragmented and insufficient for capturing continuous life trajectories. |
| Approach: | They propose a benchmark for evaluating long-term memory in AI Clone scenarios grounded in non-conversational digital traces, including diaries, social media posts, and emails, spanning one to three years. |
| Outcome: | Experiments show that existing memory benchmarks struggle in this setting, highlighting open challenges for life-grounded personalized AI. |
Towards Understanding Task-agnostic Debiasing Through the Lenses of Intrinsic Bias and Forgetfulness (2024.findings-acl)
Copied to clipboard
Guangliang Liu, Milad Afshari, Xitong Zhang, Zhiyu Xue, Avrajit Ghosh, Bidhan Bashyal, Rongrong Wang, Kristen Johnson
| Challenge: | Debiasing Pretrained Language Models (PLMs) are task-agnostic and can be generalizable, but its impact on language modeling ability and the risk of relearning social biases remain as the two most significant challenges. |
| Approach: | They propose a framework which can Propagate Socially-fair Debiasing to Downstream Fine-tuning to alleviate the forgetting issue of PLMs by regularizing debiased attention heads based on the PLM’s bias levels from stages of pretraining and debiase. |
| Outcome: | The proposed framework can Propagate Socially-fair Debiasing to Downstream Fine-tuning, indicating that the ineffectiveness of debiase can be alleviated by overcoming the forgetting issue through regularizing successfully debiased attention heads based on the PLMs’ bias levels from stages of pretraining and debiases. |
Communication-Efficient and Tensorized Federated Fine-Tuning of Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) excel in translation and summarization due to the capabilities of transformer architectures. |
| Approach: | They propose to integrate tensorized adapters into model encoder/decoder blocks to improve model adaptability against data heterogeneity. |
| Outcome: | Experiments on large-scale cross-device FL and large-silo FL show that the proposed methods perform on par or even better than existing federated PEFT approaches while reducing communication cost. |
LMR-BENCH: Evaluating LLM Agent’s Ability on Reproducing Language Modeling Research (2025.emnlp-main)
Copied to clipboard
Shuo Yan, Ruochen Li, Ziming Luo, Zimu Wang, Daoyang Li, Liqiang Jing, Kaiyu He, Peilin Wu, Juntong Ni, George Michalopoulos, Yue Zhang, Ziyang Zhang, Mian Zhang, Zhiyu Chen, Xinya Du
| Challenge: | Large language model (LLM) agents have demonstrated remarkable potential in advancing scientific discovery, but their capability in reproducing code from research papers remains underexplored. |
| Approach: | They propose to evaluate LLM agents' ability to reproduce scientific research papers by analyzing code reproduction tasks from 23 research papers published in top-tier NLP venues. |
| Outcome: | The proposed benchmark systematically evaluates the capability of large language model (LLM) agents on code reproduction from Language Modeling Research. |
QAEncoder: Towards Aligned Representation Learning in Question Answering Systems (2025.acl-long)
Copied to clipboard
Zhengren Wang, Qinhan Yu, Shida Wei, Zhiyu Li, Feiyu Xiong, Xiaoxing Wang, Simin Niu, Hao Liang, Wentao Zhang
| Challenge: | Modern QA systems entail retrieval-augmented generation (RAG) for accurate and trustworthy responses, but the inherent gap between user queries and relevant documents hinders precise matching. |
| Approach: | They propose a retrieval-augmented generation (RAG)-based approach to bridge this gap by attaching document fingerprints to the embedding to estimate the expectation of potential queries. |
| Outcome: | Experiments across diverse datasets, languages, and embedding models confirm the proposed solution is simple-yet-effective with zero additional index storage, retrieval latency, training costs, or catastrophic forgetting and hallucination issues. |
Logic2Text: High-Fidelity Natural Language Generation from Logical Forms (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies on Natural Language Generation (NLG) from structured data focus on surface descriptions of simple record sequences, for example, attribute-value pairs of fixed or very limited schema. |
| Approach: | They propose to use a large-scale dataset to generate NLG from logical forms to obtain controllable and faithful generations from structured data. |
| Outcome: | The proposed model can describe interesting facts from logical inferences across records, but it is difficult to produce such fidelity. |
IDEA: Enhancing the Rule Learning Ability of Large Language Model Agent through Induction, Deduction, and Abduction (2025.findings-acl)
Copied to clipboard
| Challenge: | RULEARN is a benchmark to assess the rule-learning abilities of large language models (LLMs) in interactive environments. |
| Approach: | They propose a framework that integrates the process of **I**nduction, **De**duction, and **A**bduction. |
| Outcome: | The proposed framework improves on the baseline and human-like rule learning in real-world scenarios. |
Search Wisely: Mitigating Sub-optimal Agentic Searches By Reducing Uncertainty (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have propelled their use in informationintensive tasks such as question answering and knowledge synthesis. |
| Approach: | They propose a reinforcement learning-based training method that incorporates confidence thresholds to reward high-certainty search decisions. |
| Outcome: | The proposed method outperforms baseline models on seven QA benchmarks and demonstrates that it is more efficient than existing methods. |
HopRAG: Multi-Hop Reasoning for Logic-Aware Retrieval-Augmented Generation (2025.findings-acl)
Copied to clipboard
| Challenge: | Traditional retrieval systems focus on lexical or semantic similarity rather than logical relevance. |
| Approach: | They propose a new RAG framework that augments retrieval with logical reasoning . hopRAG uses a retrieve-reason-prune mechanism to explore multi-hop neighbors . |
| Outcome: | The proposed framework outperforms conventional retrieval systems and state-of-the-art benchmarks on multi-hop QA tasks. |
SafeRAG: Benchmarking Security in Retrieval-Augmented Generation of Large Language Model (2025.acl-long)
Copied to clipboard
Xun Liang, Simin Niu, Zhiyu Li, Sensen Zhang, Hanyu Wang, Feiyu Xiong, Zhaoxin Fan, Bo Tang, Jihao Zhao, Jiawei Yang, Shichao Song, Mengwei Wang
| Challenge: | Existing approaches to integrating external knowledge into large language models (LLMs) however, the incorporation of external knowledge increases the vulnerability of LLMs . |
| Approach: | They propose a benchmark to evaluate the RAG security using a dataset . they classify attack tasks into silver noise, inter-context conflict, soft ad, and white Denial-of-Service . |
| Outcome: | The proposed benchmark evaluates the security of RAG against 14 representative RAG components. |
PAC-tuning: Fine-tuning Pre-trained Language Models with PAC-driven Perturbed Gradient Descent (2023.emnlp-main)
Copied to clipboard
| Challenge: | PAC-tuning is a two-stage fine-tune method for pretrained language models . PAC training minimizes the PACBayes generalization bound to learn proper parameter distribution . |
| Approach: | They propose a two-stage fine-tuning method to minimize the PAC-Bayes generalization bound . they use PAC to inject noise with variance learned in the first stage into the model parameters . |
| Outcome: | The proposed method outperforms baseline methods on 5 GLUE benchmark tasks. |