Papers by Zhihan Zhang

Pre-training Language Models for Comparative Reasoning (2023.emnlp-main)

Challenge:	Existing methods for retrieving encyclopedic knowledge lack a large corpus and effective commonsense retriever.
Approach:	They propose a framework for retrieval-augmented commonsense reasoning with a large commonsensense corpus and a commonseense retriever.
Outcome:	The proposed framework outperforms existing methods on commonsense reasoning tasks.

Copied to clipboard

Mengxia Yu, Zhihan Zhang, Wenhao Yu, Meng Jiang

Cross-Modal Commentator: Automatic Machine Commenting Based on Cross-Modal Information (P19-1)

Challenge:	Comparative reasoning is a process of comparing objects, concepts, or entities to draw conclusions.
Approach:	They propose a framework to pre-train language models for enhancing comparative reasoning abilities . they collect scalable data for text-based entity comparison .
Outcome:	The proposed framework significantly improves comparative reasoning abilities under low-resource conditions on downstream tasks.

Copied to clipboard

Pengcheng Yang, Zhihan Zhang, Fuli Luo, Lei Li, Chengyang Huang, Xu Sun

Aligning Large Language Models with Implicit Preferences from User-Generated Content (2025.acl-long)

Challenge:	Existing work on commenting based on textual content is focused on other modalities, such as graphics and images.
Approach:	They propose a task to integrate multiple modalities into automatic commenting . they construct a large-scale dataset and propose 'co-attention' model to capture dependency between textual and visual information.
Outcome:	The proposed model can achieve better performance than baselines.

Copied to clipboard

Zhaoxuan Tan, Zheng Li, Tianyi Liu, Haodong Wang, Hyokun Yun, Ming Zeng, Pei Chen, Zhihan Zhang, Yifan Gao, Ruijie Wang, Priyanka Nigam, Bing Yin, Meng Jiang

ALERT: An LLM-powered Benchmark for Automatic Evaluation of Recommendation Explanations (2025.naacl-long)

Challenge:	Existing preference learning methods rely heavily on curated data from humans or advanced LLMs, which is costly and difficult to scale.
Approach:	They propose a framework that leverages implicit preferences in unlabeled user-generated content to generate preference data.
Outcome:	The proposed framework transforms user-generated content into user queries and generates responses from the policy model.

Copied to clipboard

Yichuan Li, Xinyang Zhang, Chenwei Zhang, Mao Li, Tianyi Liu, Pei Chen, Yifan Gao, Kyumin Lee, Kaize Ding, Zhengyang Wang, Zhihan Zhang, Jingbo Shang, Xian Li, Trishul Chilimbi

Responsible & Trustworthy NLP

Challenge:	Existing benchmarks for recommendation explanation evaluation lack item diversity and user preferences data.
Approach:	They propose a model-agnostic recommendation explanation evaluation benchmark based on Amazon e-commerce categories with implicit preferences . they propose two novel automatic evaluators that enable scalable and human-preference aligned evaluation of explanations .
Outcome:	The proposed model-agnostic evaluation benchmark outperforms existing methods in a variety of domains.

Diversifying Content Generation for Commonsense Reasoning with Mixture of Knowledge Graph Experts (2022.findings-acl)

Copied to clipboard

Wenhao Yu, Chenguang Zhu, Lianhui Qin, Zhihan Zhang, Tong Zhao, Meng Jiang

Learning Dialogue Representations from Consecutive Utterances (2022.naacl-main)

Challenge:	Recent years have seen a surge of interest in improving the generation quality of commonsense reasoning tasks.
Approach:	They propose a method that diversifies the generative reasoning by a mixture of expert strategy on commonsense knowledge graphs to encourage various generation outputs.
Outcome:	The proposed method improves diversity while achieving on par performance on two GCR benchmarks, based on both automatic and human evaluations.

Copied to clipboard

Zhihan Zhou, Dejiao Zhang, Wei Xiao, Nicholas Dingwall, Xiaofei Ma, Andrew Arnold, Bing Xiang

Semantic Text Processing

Challenge:	Dialogue Sentence Embedding (DSE) is a self-supervised contrastive learning method that learns effective dialogue representations suitable for a wide range of dialogue-oriented tasks.
Approach:	They propose a self-supervised contrastive learning method that learns dialogue representations suitable for a wide range of dialogue tasks.
Outcome:	The proposed method outperforms baselines on five dialogue tasks on a few-shot and zero-shot datasets.

IHEval: Evaluating Language Models on Following the Instruction Hierarchy (2025.naacl-long)

Copied to clipboard

Zhihan Zhang, Shiyang Li, Zixuan Zhang, Xin Liu, Haoming Jiang, Xianfeng Tang, Yifan Gao, Zheng Li, Haodong Wang, Zhaoxuan Tan, Yichuan Li, Qingyu Yin, Bing Yin, Meng Jiang

AutoRubric: Rubric-Based Generative Rewards for Faithful Multimodal Reasoning (2026.findings-acl)

Challenge:	Instruction-tuned language models (LMs) are increasingly deployed as interactive services across various applications.
Approach:	They propose a benchmark to evaluate models' ability to follow the instruction hierarchy by comparing their models to a set of benchmarks.
Outcome:	The proposed benchmark covers 3,538 examples across nine tasks covering cases where instructions in different priorities either align or conflict.

Copied to clipboard

Mengzhao Jia, Zhihan Zhang, Ignacio Cases, Zheyuan Liu, Meng Jiang, Peng Qi

TOWER: Tree Organized Weighting for Evaluating Complex Instructions (2024.findings-emnlp)

Challenge:	Multimodal large language models (MLLMs) have advanced from perception tasks to complex multi-step reasoning.
Approach:	They propose a framework that integrates reinforcement learning with verifiable rewards with process-level supervision through automatically collected rubric-based generative rewards.
Outcome:	The proposed framework achieves state-of-the-art performance on six multimodal reasoning benchmarks and significantly improves reasoning faithfulness in dedicated evaluations.

Copied to clipboard

Noah Ziems, Zhihan Zhang, Meng Jiang

Challenge:	Evaluating the ability of large language models to follow human-written instructions remains a challenge.
Approach:	They propose a new evaluation metric that incorporates human-judged importance into the assessment of complex instruction following.
Outcome:	The proposed evaluation metric incorporates human-judged importance into the assessment of complex instruction following.

Advancing Language Models through Instruction Tuning: Recent Progress and Challenges (2025.emnlp-tutorials)

Copied to clipboard

Zhihan Zhang, Renze Lou, Fangkai Jiao, Wenpeng Yin, Meng Jiang

Enhancing Mathematical Reasoning in LLMs by Stepwise Correction (2025.acl-long)

Challenge:	tutorial addresses three critical questions within the field of instruction tuning: (1) What are the current focal points in instruction tuning research? (2) What are best practices in training an instruction-following model? (3) What new challenges have emerged?
Approach:	This tutorial presents a systematic overview of recent advances in instruction tuning.
Outcome:	The tutorial covers different stages in model training: supervised fine-tuning, preference optimization, and reinforcement learning.

Copied to clipboard

Zhenyu Wu, Qingkai Zeng, Zhihan Zhang, Zhaoxuan Tan, Chao Shen, Meng Jiang

A Survey of Multi-task Learning in Natural Language Processing: Regarding Task Relatedness and Training Methods (2023.eacl-main)

Challenge:	Existing Best-of-N decoding methods often lead to incorrect solutions . a novel method is proposed to help large language models identify and revise incorrect steps in their generated reasoning paths.
Approach:	They propose a method that helps large language models identify and revise incorrect steps in their generated reasoning paths.
Outcome:	The proposed method outperforms the state-of-the-art Best-ofN decoding method by +2.4 and reduces token consumption by 77.8%.

Copied to clipboard

Zhihan Zhang, Wenhao Yu, Mengxia Yu, Zhichun Guo, Meng Jiang

Semantic Text Processing

Challenge:	Multi-task learning is a popular approach in natural language processing because of its commonalities and differences.
Approach:	They propose to summarize recent advances in multi-task learning methods based on their task relatedness into two general multi-step training methods.
Outcome:	The proposed methods summarize the tasks and discuss future directions.

Toward Optimal LLM Alignments Using Two-Player Games (2025.findings-emnlp)

Copied to clipboard

Rui Zheng, Hongyi Guo, Zhihan Liu, Xiaoying Zhang, Yuanshun Yao, Xiaojun Xu, Zhaoran Wang, Zhiheng Xi, Tao Gui, Qi Zhang, Xuanjing Huang, Yang Liu, Hang Li

Responsible & Trustworthy NLP

Challenge:	Alignment of large language models (LLM) is a process that ensures the model’s responses to user prompts align with human intentions and social values.
Approach:	They propose an alignment method based on a two-agent game consisting of an adversarial agent and a defensive agent.
Outcome:	The proposed method improves on a two-agent game with an adversarial agent and a defensive agent.

MultiChartQA: Benchmarking Vision-Language Models on Multi-Chart Problems (2025.naacl-long)

Copied to clipboard

Zifeng Zhu, Mengzhao Jia, Zhihan Zhang, Lang Li, Meng Jiang

Auto-Instruct: Automatic Instruction Generation and Ranking for Black-Box Language Models (2023.findings-emnlp)

Challenge:	Existing chart understanding benchmarks focus on single-chart tasks, neglecting multi-hop reasoning required to extract and integrate information from multiple charts.
Approach:	They propose a benchmark that evaluates MLLMs’ capabilities in four key areas: direct question answering, parallel question answering and comparative reasoning.
Outcome:	The proposed benchmark evaluates MLLMs’ capabilities in four key areas: direct question answering, parallel question answering and comparative reasoning.

Copied to clipboard

Zhihan Zhang, Shuohang Wang, Wenhao Yu, Yichong Xu, Dan Iter, Qingkai Zeng, Yang Liu, Chenguang Zhu, Meng Jiang

Large Language Models are Built-in Autoregressive Search Engines (2023.findings-acl)

Challenge:	Large language models can perform a wide range of tasks by following natural language instructions without task-specific fine-tuning.
Approach:	They propose a method to automatically improve the quality of LLM instructions . they leverage the generative ability of LMS to generate diverse candidate instructions based on a scoring model trained on 575 existing NLP tasks.
Outcome:	The proposed method surpasses human-written and LLM-generated instructions on 118 out-of-domain tasks.

Copied to clipboard

Noah Ziems, Wenhao Yu, Zhihan Zhang, Meng Jiang

Information Retrieval

Challenge:	Existing dual-encoder dense retrievers obtain representations for questions and documents independently, allowing only shallow interactions between them.
Approach:	They propose to use large language models to generate URLs for document retrieval by following human instructions.
Outcome:	The proposed method achieves better retrieval performance than existing retrieval approaches on open-domain question answering benchmarks.

GRAG: Graph Retrieval-Augmented Generation (2025.findings-naacl)

Copied to clipboard

Yuntong Hu, Zhihan Lei, Zheng Zhang, Bo Pan, Chen Ling, Liang Zhao

Structured Data in NLP

Challenge:	Naive Retrieval-Augmented Generation (RAG) focuses on individual documents during retrieval and is not suitable for networked documents.
Approach:	They propose a novel divide-and-conquer strategy that retrieves optimal subgraph structure in linear time.
Outcome:	The proposed approach outperforms current state-of-the-art methods on graph reasoning benchmarks.

MMTutorBench: The First Multimodal Benchmark for AI Math Tutoring (2026.acl-long)

Copied to clipboard

Tengchao Yang, Sichen Guo, Mengzhao Jia, Jiaming Su, Yuanyang Liu, Zhihan Zhang, Meng Jiang

Numerical Reasoning

Challenge:	Existing benchmarks for AI math tutoring largely overlook these skills.
Approach:	They evaluate 12 leading multimodal large language models and find clear performance gaps between them.
Outcome:	The proposed benchmarks show that they can solve 770 problems and provide diagnostics and guidance to students step by step.

PLUG: Leveraging Pivot Language in Cross-Lingual Instruction Tuning (2024.acl-long)

Copied to clipboard

Zhihan Zhang, Dong-Ho Lee, Yuwei Fang, Wenhao Yu, Mengzhao Jia, Meng Jiang, Francesco Barbieri

Cross-Lingual Transfer

Challenge:	Instruction tuning has advanced large language models (LLMs) but its application in lower-resource languages faces challenges due to the imbalanced foundational abilities of LLMs across different languages.
Approach:	They propose a pivot language guided generation approach that utilizes a high-resource language as the pivot to enhance instruction tuning in lower-resourced languages.
Outcome:	The proposed approach improves instruction-following abilities of LLMs by 29% on average compared to directly responding in the target language alone.

Exploring Contrast Consistency of Open-Domain Question Answering Systems on Minimally Edited Questions (2023.tacl-1)

Copied to clipboard

Zhihan Zhang, Wenhao Yu, Zheng Ning, Mingxuan Ju, Meng Jiang

Natural Language Interfaces

Challenge:	Existing studies on the ability of a model to make consistently correct predictions in the presence of perturbations have not been conducted in open-domain question answering (OpenQA).
Approach:	They propose a query-side contrastive loss to improve the dense passage retriever (DPR) to improve DPR training.
Outcome:	The proposed approach improves the density of the dense passage retriever (DPR) training set without sacrificing accuracy on standard test sets.

Large Language Models Can Self-Correct with Key Condition Verification (2024.emnlp-main)

Copied to clipboard

Zhenyu Wu, Qingkai Zeng, Zhihan Zhang, Zhaoxuan Tan, Chao Shen, Meng Jiang

Analyzing Temporal Complex Events with Large Language Models? A Benchmark towards Temporal, Long Context Understanding (2024.acl-long)

Challenge:	Existing methods to correct reasoning without external feedback have not been used in large language models.
Approach:	They propose an iterative verify-then-correct framework to progressively identify and correct (probably) false responses, named ProCo.
Outcome:	The proposed method improves the accuracy of LLMs on three reasoning tasks.

Copied to clipboard

Zhihan Zhang, Yixin Cao, Chenchen Ye, Yunshan Ma, Lizi Liao, Tat-Seng Chua

Semantic Text Processing

Challenge:	Existing research in complex event analysis has made significant strides but is constrained by inadequate natural language processing techniques.
Approach:	They propose a novel approach using Large Language Models to extract and analyze the event chain within TCE, characterized by their key points and timestamps.
Outcome:	The proposed model performs comparable to models with long context window and retrieval-augmented generation method in three distinct tasks .

From General Reward to Targeted Reward: Improving Open-ended Long-context Generation Models (2025.emnlp-main)

Copied to clipboard

Zhihan Guo, Jiele Wu, Wenqian Cui, Yifei Zhang, Minda Hu, Yufei Wang, Irwin King

Text Generation

Challenge:	Current research on long-form context in Large Language Models (LLMs) focuses on understanding of long-contexts, but the open-ended Long Text Generation (Open-LTG) remains underexplored.
Approach:	They propose a method that uses data synthesis and a reward signal to enhance model performance.
Outcome:	The proposed method outperforms GPT-4-Turbo and improves performance by 20% on the Open-LTG task.

A Unified Encoder-Decoder Framework with Entity Memory (2022.emnlp-main)

Copied to clipboard

Zhihan Zhang, Wenhao Yu, Chenguang Zhu, Meng Jiang

Learn Beyond The Answer: Training Language Models with Reflection for Mathematical Reasoning (2024.emnlp-main)

Challenge:	Existing approaches to index, retrieve, and read documents as evidence suffer from large computational overheads.
Approach:	They propose an encoder-decoder framework with an entity memory that stores entity knowledge as latent representations and pre-trained on Wikipedia along with encoder parameters.
Outcome:	The proposed framework outperforms memory-based and non-memory encoder-decoder models on various entity-intensive question answering and generation tasks.

Copied to clipboard

Zhihan Zhang, Tao Ge, Zhenwen Liang, Wenhao Yu, Dian Yu, Mengzhao Jia, Dong Yu, Meng Jiang

Detecting RAG Extraction Attack via Dual-Path Runtime Integrity Game (2026.acl-long)

Challenge:	Existing studies focus on broadening the training set with data augmentation techniques to maximize such benefits.
Approach:	They propose a method that embeds problem reflection into each training instance.
Outcome:	The proposed method enhances performance in standard and complex scenarios that require reflective thinking.

Copied to clipboard

Yuanbo Xie, Yingjie Zhang, Yulin Li, Shouyou Song, Xiaokun Chen, Zhihan Liu, Liya Su, Tingwen Liu

Robustness in NLP

Challenge:	Retrieval-Augmented Generation (RAG) systems augment large language models with external knowledge, but introduce a critical security vulnerability: Knowledge Base Leakage.
Approach:	They propose a runtime defense mechanism inspired by stack canaries in software security . canaryRAG embeds carefully designed canary tokens into retrieved chunks and reformulates RAG extraction defense as a dual-path runtime integrity game.
Outcome:	The proposed system can detect and prevent RAG Knowledge Base Leakage in real time . it can be integrated into arbitrary RAG pipelines without retraining or structural modifications .

XFinBench: Benchmarking LLMs in Complex Financial Problem Solving and Reasoning (2025.findings-acl)

Copied to clipboard

Zhihan Zhang, Yixin Cao, Lizi Liao