Papers by Bo Su
CE-RM: A Pointwise Generative Reward Model Optimized via Two-Stage Rollout and Unified Criteria (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have shown that rule-based evaluation methods are ineffective for open-ended natural language generation. |
| Approach: | They propose a pointwise generative reward model with a dedicated two-stage rollout method and unified query-based criteria that can be trained with 5.7K high-quality data. |
| Outcome: | The proposed model achieves superior performance on diverse reward model benchmarks, especially in Best-of-N scenarios, and delivers more effective improvements in downstream RL practice. |
Chinese SafetyQA: A Safety Short-form Factuality Benchmark for Large Language Models (2025.acl-long)
Copied to clipboard
Yingshui Tan, Boren Zheng, Baihui Zheng, Kerui Cao, Huiyun Jing, Jincheng Wei, Jiaheng Liu, Yancheng He, Wenbo Su, Xiaoyong Zhu, Bo Zheng, Kaifu Zhang
| Challenge: | Large language models have created significant safety concerns . factuality ability is crucial in determining whether they can be deployed and applied safely and compliantly within specific regions. |
| Approach: | They propose a benchmark to evaluate the factuality of large language models in China . they evaluate the models' ability to provide accurate and reliable information . |
| Outcome: | The proposed benchmark evaluates the factuality abilities of existing LLMs and compares them to LLM abilities. |
E2-LLM: Efficient and Extreme Length Extension of Large Language Models (2024.findings-acl)
Copied to clipboard
Jiaheng Liu, ZhiqiBai ZhiqiBai, Yuanxing Zhang, Chenchen Zhang, YuangZh YuangZh, Ge Zhang, JiakaiWang JiakaiWang, Haoran Que, Yukang Chen, Wenbo Su, Tiezheng Ge, Jie Fu, Wenhu Chen, Bo Zheng
| Challenge: | Existing techniques for extending context capabilities in LLMs require additional training procedures and access to datasets with long context (e.g., sequences of 32K tokens). |
| Approach: | They propose a solution to extend context capabilities in Large Language Models by training a single process over a sequence of 4K tokens. |
| Outcome: | The proposed solution significantly reduces the cost of continual-pretraining or fine-tuning over short sequences and improves robustness to diverse relative positions. |
USB: A COMPREHENSIVE AND UNIFIED SAFETY EVALUATION BENCHMARK FOR MULTIMODAL LARGE LANGUAGE MODELS (2026.acl-long)
Copied to clipboard
Baolin Zheng, Guanlin Chen, Qingyang Teng, Hongqiong Zhong, Yingshui Tan, Zhendong Liu, Weixun Wang, Jiaheng Liu, Jian Yang, Huiyun Jing, Jincheng Wei, Wenbo Su, Xiaoyong Zhu, Bo Zheng, Kaifu Zhang
| Challenge: | Existing safety benchmarks fail to provide reliable assessments due to limited risk coverage, insufficient scale and the oversight of complex modality combinations. |
| Approach: | They propose a framework that covers 61 risk categories across four modality interactions to address this gap. |
| Outcome: | The proposed framework covers 61 risk categories across four distinct modality interactions. |
Correctable-DST: Mitigating Historical Context Mismatch between Training and Inference for Improved Dialogue State Tracking (2022.emnlp-main)
Copied to clipboard
Hongyan Xie, Haoxiang Su, Shuangyong Song, Hao Huang, Bo Zou, Kun Deng, Jianghua Lin, Zhihui Zhang, Xiaodong He
| Challenge: | Existing dialogue state tracking approaches predict the dialogue state of a target turn sequentially based on the ground-truth previous dialogue state. |
| Approach: | They propose a method that predicts dialogue state sequentially based on previous dialogue state . they propose generating a previously “predicted” dialogue state using ground-truth previous dialogue states . |
| Outcome: | The proposed method achieves 67.51%, 68.24%, 70.30%, 71.38%, and 81.27% joint goal accuracy on MultiWOZ 2.0-2.4 datasets. |
SELECting over Tokens: Curating Pre-training Data at Scale via Token Classification (2026.acl-long)
Copied to clipboard
Xin Tong, Weidong Zhang, Jiaang Li, Haibin Chen, Shilei Liu, Langming Liu, Kangtao Lv, Yujin Yuan, Wenbo Su, Bo Zheng
| Challenge: | Existing pipelines rely on expert-crafted heuristic rules, which lack content-aware, fine-grained noise detection. |
| Approach: | They propose a framework that reframes data refinement as a highly efficient token classification task. |
| Outcome: | The proposed framework outperforms existing pipelines on benchmarks and is 2.5x faster at inference. |
Beyond Black-Box Interventions: Latent Probing for Faithful Retrieval-Augmented Generation (2026.findings-acl)
Copied to clipboard
Linfeng Gao, Qinggang Zhang, Baolong Bi, Bo Zeng, Zheng Yuan, Zerui Chen, Zhimin Wei, Shenghua Liu, Linlong Xu, Longyue Wang, Weihua Luo, Jinsong Su
| Challenge: | Existing approaches to improve contextual faithfulness treat the LLM as a black box, generating responses that are inconsistent with the provided context. |
| Approach: | They propose a framework for faithful RAG that operates in three stages: (i) fine-grained knowledge pruning to filter irrelevant context, (ii) latent conflict probing to identify hard conflicts in the model’s latent space, and (iv) conflict-aware attention to modulate attention heads toward faithful context integration. |
| Outcome: | Experiments show that ProbeRAG significantly improves both accuracy and contextual faithfulness. |
ShopSimulator: Evaluating and Exploring RL-Driven LLM Agent for Shopping Assistants (2026.acl-long)
Copied to clipboard
Pei Wang, Yanan Wu, Xiaoshuai Song, Weixun Wang, Gengru Chen, Zhongwen Li, Kezhong Yan, Qi Liu, Ken Deng, Shuaibing Zhao, Shaopan Xiong, Xuepeng Liu, Xuefeng Chen, Wanxi Deng, Wenbo Su, Bo Zheng
| Challenge: | Existing studies on large language model-based agents focus on evaluation benchmarks without training support. |
| Approach: | They propose a large-scale Chinese shopping simulation environment that uses large language models to train agents. |
| Outcome: | The proposed model performs poorly in a large-scale and challenging shopping environment in China. |
Read As Human: Compressing Context via Parallelizable Close Reading and Skimming (2026.acl-long)
Copied to clipboard
Jiwei Tang, Shilei Liu, Zhicheng Zhang, Qingsong Lv, Runsong Zhao, Tingwei Lu, Langming Liu, Haibin Chen, Yujin Yuan, Hai-Tao Zheng, Wenbo Su, Bo Zheng
| Challenge: | Existing task-aware methods require loading the entire input sequence at once for compression, which suffer from computational inefficiency. |
| Approach: | They propose a framework that adopts an adaptive hybrid reading strategy to reduce computational inefficiency and redundant information in long-context scenarios. |
| Outcome: | Experiments show that RAM outperforms baselines on multiple question answering and summarization benchmarks while delivering up to a 12x speedup on long inputs. |
See the World, Discover Knowledge: A Chinese Factuality Evaluation for Large Vision Language Models (2025.findings-acl)
Copied to clipboard
Jihao Gu, Yingyao Wang, Pi Bu, Chen Wang, Ziming Wang, Tengtao Song, Donglai Wei, Jiale Yuan, Yingxiu Zhao, Yancheng He, Shilong Li, Jiaheng Liu, Meng Cao, Jun Song, Yingshui Tan, Xiang Li, Wenbo Su, Xiaoyong Zhu, Bo Zheng
| Challenge: | Existing models for large vision language models do not fully reflect their knowledge capacity and reliability, resulting in erroneous outputs that do not align with the image content or provide answers lacking knowledge evidence. |
| Approach: | They propose a Chinese-based benchmark for visual factuality across 8 major topics and 56 subtopics and a multi-hop question construction. |
| Outcome: | The proposed model decouples visual factuality into two parts: seeing the world and discovering knowledge. |
How to inject knowledge efficiently? Knowledge Infusion Scaling Law for Pre-training Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies show that strategically infusing domain knowledge during pretraining can substantially improve downstream performance. |
| Approach: | They propose a knowledge infusion scaling law that predicts the optimal amount of domain knowledge to inject into large LLMs by analyzing their smaller counterparts. |
| Outcome: | The proposed model predicts the optimal amount of domain knowledge to inject into large LLMs by analyzing their smaller counterparts. |
Entity-Aware Abstractive Multi-Document Summarization (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing models for multidocument summarization do not focus on explicitly modeling the underlying semantic information across documents. |
| Approach: | They propose an entityaware model for abstractive multi-document summarization that augments the classical Transformer-based encoder-decoder framework with a heterogeneous graph consisting of text units and entities as nodes. |
| Outcome: | The proposed model can deal with saliency and redundancy issues explicitly and can be used with pre-trained language models, arriving at improved performance. |
2D-DPO: Scaling Direct Preference Optimization with 2-Dimensional Supervision (2025.findings-naacl)
Copied to clipboard
Shilong Li, Yancheng He, Hui Huang, Xingyuan Bu, Jiaheng Liu, Hangyu Guo, Weixun Wang, Jihao Gu, Wenbo Su, Bo Zheng
| Challenge: | Existing methods that optimize for scalar scores or ranking reward ignore multi-dimensional nature of human preferences. |
| Approach: | They propose to extend the preference of Direct Preference Optimization to two dimensions: segments and aspects. |
| Outcome: | The proposed framework decomposes the overall objective into multi-segment and multi-aspect objectives. |
OPeRA: A Dataset of Observation, Persona, Rationale, and Action for Evaluating LLMs on Human Online Shopping Behavior Simulation (2026.acl-long)
Copied to clipboard
Ziyi Wang, Yuxuan Lu, Wenbo Li, Amirali Amini, Bo Sun, Yakov Bart, Weimin Lyu, Jiri Gesi, Tian Wang, Jing Huang, Yu Su, Upol Ehsan, Malihe Alikhani, Toby Jia-Jun Li, Lydia Chilton, Dakuo Wang
| Challenge: | evaluating LLMs' ability to mimic real user behavior remains an open challenge due to the lack of high-quality, publicly available datasets that capture both the observable actions and the internal reasoning of an actual user. |
| Approach: | They propose a dataset of Observation, Persona, Rationale, and Action collected from real human participants during online shopping sessions. |
| Outcome: | The proposed dataset is the first to evaluate how well current LLMs can accurately simulate the next web action of a specific user. |
AIR: Complex Instruction Generation via Automatic Iterative Refinement (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for generating complex instructions are resource-intensive and lack diversity. |
| Approach: | They propose a framework to generate complex instructions with constraints using a document-generated initial instruction and an iterative refinement framework to incorporate LLM-as-judge guidance. |
| Outcome: | The proposed framework significantly outperforms existing methods for generating complex instructions, and outperformed existing methods. |
RAVR: Reference-Answer-guided Variational Reasoning for Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Experiments show that reinforcement learning (RL) can refine the reasoning abilities of large language models (LLMs) but requires a key prerequisite: the model must already be able to generate high-utility reasoning paths with non-negligible probability. |
| Approach: | They propose a framework that uses answer-conditioned reasoning as a variational surrogate for question-only reasoning. |
| Outcome: | Experiments on 11 benchmarks and 3 models show that RAVR reduces hesitation, strengthens conclusion consolidation, and promotes problem-specific strategies in reasoning. |
Targeting the Needle, Ignoring the Haystack: Anchoring Crucial Cues for Evolving Scam Call Detection via an LLM-Assisted Classifier (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for fraud detection on online service platforms often fail to generalize due to the scarcity of labeled data and the continuous evolution of conversational contexts. |
| Approach: | They propose a framework that anchors detection on Semantic Primitives . they prioritize stable evidence over conversational noise to ensure a verifiable fraud tactic . |
| Outcome: | The proposed framework achieves superior robustness and efficiency compared to baselines . it prioritizes stable evidence over diverse conversational noise . |
CoMeT: Collaborative Memory Transformer for Efficient Long Context Modeling (2026.acl-long)
Copied to clipboard
Runsong Zhao, Shilei Liu, Jiwei Tang, Langming Liu, Haibin Chen, Weidong Zhang, Yujin Yuan, Tong Xiao, JingBo Zhu, Wenbo Su, Bo Zheng
| Challenge: | a novel architecture that enables LLMs to handle arbitrarily long sequences with constant memory usage and linear time complexity is a major barrier to long-context processing. |
| Approach: | They propose a novel architecture that enables LLMs to handle arbitrarily long sequences with constant memory usage and linear time complexity. |
| Outcome: | The proposed architecture can handle arbitrarily long sequences with constant memory usage and linear time complexity. |
Entity-centered Cross-document Relation Extraction (2022.emnlp-main)
Copied to clipboard
Fengqi Wang, Fei Li, Hao Fei, Jingye Li, Shengqiong Wu, Fangfang Su, Wenxuan Shi, Donghong Ji, Bo Cai
| Challenge: | Existing methods for relation extraction only use text snippets surrounding target entities in multiple documents. |
| Approach: | They propose a relation-extraction model that uses cross-path entity relation attention to detect the semantic relations between entities in a given text. |
| Outcome: | The proposed method outperforms the state-of-the-art methods in the dataset CodRED by 10%. |
Do Vision-Language Models Have Internal World Models? Towards an Atomic Evaluation (2025.findings-acl)
Copied to clipboard
Qiyue Gao, Xinyu Pi, Kevin Liu, Junrong Chen, Ruolan Yang, Xinqi Huang, Xinyu Fang, Lu Sun, Gautham Kishore, Bo Ai, Stone Tao, Mengyang Liu, Jiaxi Yang, Chao-Jung Lai, Chuanyang Jin, Jiannan Xiang, Benhao Huang, Zeming Chen, David Danks, Hao Su, Tianmin Shu, Ziqiao Ma, Lianhui Qin, Zhiting Hu
| Challenge: | Recent studies have evaluated and shown limitations in specific capabilities such as visual understanding, but a systematic evaluation of VLMs’ fundamental WM abilities remains absent. |
| Approach: | They propose a framework that assesses perception and prediction to provide an atomic evaluation of VLMs as WMs. |
| Outcome: | The proposed framework assesses perception and prediction abilities on 15 latest VLMs and compares them to human-level models. |
Naive Bayes-based Context Extension for Large Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have shown promising in-context learning abilities, but are often impeded by length limitations of transformer architecture. |
| Approach: | They propose a framework to enable existing Large Language Models to perform ICL with an increased number of demonstration examples by significantly expanding their context size. |
| Outcome: | The proposed framework outperforms existing methods in in-context learning with an increased number of demonstration examples while preserving linear efficiency. |
PretrainRL: Alleviating Factuality Hallucination of Large Language Models at the Beginning (2026.findings-acl)
Copied to clipboard
Langming Liu, Kangtao Lv, Haibin Chen, Weidong Zhang, Yejing Wang, Shilei Liu, Xin Tong, Yujin Yuan, Yongwei Wang, Wenbo Su, Bo Zheng
| Challenge: | Large language models suffer from factual hallucinations where they generate verifiable falsehoods. |
| Approach: | They propose a framework that integrates reinforcement learning into the pretraining phase to consolidate factual knowledge. |
| Outcome: | The proposed framework significantly alleviates factual hallucinations and outperforms state-of-the-art methods. |
RTCFake: Speech Deepfake Detection in Real-Time Communication (2026.findings-acl)
Copied to clipboard
Jun Xue, Zhuolin Yi, Yihuan Huang, Yanzhen Ren, Yujie Chen, Cunhang Fan, Zicheng Su, Yongcheng Zhang, Bo Cai
| Challenge: | Existing detection studies focus on offline simulations and struggle to cope with complex distortions introduced during RTC transmission. |
| Approach: | They propose a large-scale speech deepfake dataset tailored for RTC scenarios . the dataset is constructed by transmitting speech through multiple social media and conferencing platforms . |
| Outcome: | The proposed dataset is constructed by transmitting speech through multiple mainstream social media and conferencing platforms, enabling precise pairing between offline and online speech. |
Chinese SimpleQA: A Chinese Factuality Evaluation for Large Language Models (2025.acl-long)
Copied to clipboard
Yancheng He, Shilong Li, Jiaheng Liu, Yingshui Tan, Weixun Wang, Hui Huang, Xingyuan Bu, Hangyu Guo, Chengwei Hu, Boren Zheng, Zhuoran Lin, Dekai Sun, Zhicheng Zheng, Wenbo Su, Bo Zheng
| Challenge: | Current frontier models sometimes generate false outputs or answers that are not substantiated by evidence. |
| Approach: | They propose Chinese SimpleQA, a Chinese benchmark to evaluate LLMs' factuality . they focus on Chinese language over 6 major topics with 99 diverse subtopics . |
| Outcome: | The Chinese SimpleQA benchmark evaluates the factuality ability of LLMs . the questions and answers are short and easy-to-evaluate . |
ConceptMath: A Bilingual Concept-wise Benchmark for Measuring Mathematical Reasoning of Large Language Models (2024.findings-acl)
Copied to clipboard
Yanan Wu, Jie Liu, Xingyuan Bu, Jiaheng Liu, Zhanhui Zhou, Yuanxing Zhang, Chenchen Zhang, ZhiqiBai ZhiqiBai, Haibin Chen, Tiezheng Ge, Wanli Ouyang, Wenbo Su, Bo Zheng
| Challenge: | ConceptMath evaluates concept-wise mathematical reasoning of Large Language Models (LLMs) Existing benchmarks that evaluate general mathematical reasoning with an average accuracy fail to probe the fine-grained failure modes of mathematical reasoning on specific datasets. |
| Approach: | They introduce a bilingual, fine-grained benchmark that evaluates concept-wise mathematical reasoning of Large Language Models. |
| Outcome: | The proposed benchmarks evaluate concept-wise mathematical reasoning of Large Language Models with concept-based accuracies. |
M2RC-EVAL: Massively Multilingual Repository-level Code Completion Evaluation (2025.acl-long)
Copied to clipboard
Jiaheng Liu, Ken Deng, Congnan Liu, Jian Yang, Shukai Liu, He Zhu, Peng Zhao, Linzheng Chai, Yanan Wu, JinKe JinKe, Ge Zhang, Zekun Moore Wang, Guoan Zhang, Yingshui Tan, Bangyu Xiang, Zhaoxiang Zhang, Wenbo Su, Bo Zheng
| Challenge: | Existing repository-level code completion benchmarks focus on a limited number of languages . existing benchmarks report overall average scores of different languages ignoring fine-grained abilities . |
| Approach: | They propose to use repository-level code completion benchmarks to evaluate general code intelligence abilities across languages for existing code Large Language Models. |
| Outcome: | The proposed benchmarks improve the code completion abilities of existing LLMs by using two types of annotations on the parsed syntax tree. |
MT-Bench-101: A Fine-Grained Benchmark for Evaluating Large Language Models in Multi-Turn Dialogues (2024.acl-long)
Copied to clipboard
Ge Bai, Jie Liu, Xingyuan Bu, Yancheng He, Jiaheng Liu, Zhanhui Zhou, Zhuoran Lin, Wenbo Su, Tiezheng Ge, Bo Zheng, Wanli Ouyang
| Challenge: | Large Language Models (LLMs) have greatly enhanced dialogue systems, but evaluation of their capabilities remains a challenge. |
| Approach: | They propose a model to evaluate the fine-grained abilities of Large Language Models in multi-turn dialogues. |
| Outcome: | The proposed model evaluates 21 popular chatbots based on MT-Bench-101 . it includes 3 overarching abilities and 13 distinct tasks within multi-turn dialogue scenarios. |
GraphReader: Building Graph-based Agent to Enhance Long-Context Abilities of Large Language Models (2024.findings-emnlp)
Copied to clipboard
Shilong Li, Yancheng He, Hangyu Guo, Xingyuan Bu, Ge Bai, Jie Liu, Jiaheng Liu, Xingwei Qu, Yangguang Li, Wanli Ouyang, Wenbo Su, Bo Zheng
| Challenge: | Existing models for long contexts struggle to handle long inputs due to limited context window and memory usage. |
| Approach: | They propose a graph-based agent system that analyzes long texts into a graphical graph . GraphReader consistently outperforms GPT-4-128k across context lengths from 16k to 256k . |
| Outcome: | The proposed model outperforms existing models on four challenging benchmarks. |
Rank-Aware Negative Training for Semi-Supervised Text Classification (2023.tacl-1)
Copied to clipboard
| Challenge: | Semi-supervised text classification-based paradigms employ the spirit of self-training, but the accuracy of pseudo-labels can be a problem in real-world scenarios. |
| Approach: | They propose a Rank-aware Negative Training framework to address SSTC in noisy label learning . they rank unlabeled texts based on evidential support from the labeled texts. |
| Outcome: | The proposed framework overcomes state-of-the-art alternatives and achieves competitive performance in other scenarios. |
ProgCo: Program Helps Self-Correction of Large Language Models (2025.acl-short)
Copied to clipboard
| Challenge: | Existing LLMs fail to self-correct and generate correct feedback, leading to misleading refinement and failure of self-refinement. |
| Approach: | They propose a program-driven self-correction approach that uses program-based verification to self-refine initial responses without external feedback. |
| Outcome: | The proposed model achieves self-correction and can be further enhanced when combined with real program tools. |
Can Large Language Models Detect Errors in Long Chain-of-Thought Reasoning? (2025.acl-long)
Copied to clipboard
Yancheng He, Shilong Li, Jiaheng Liu, Weixun Wang, Xingyuan Bu, Ge Zhang, Z.y. Peng, Zhaoxiang Zhang, Zhicheng Zheng, Wenbo Su, Bo Zheng
| Challenge: | Recent advances in o1-like models have generated long Chain-of-Thought reasoning steps to improve the reasoning abilities of existing Large Language Models (LLMs). |
| Approach: | They propose a DeltaBench to analyze the quality and effectiveness of o1-like models and measure their ability to detect errors in long COT reasoning. |
| Outcome: | The proposed model can detect errors in long COT reasoning. |
Profanity-Avoiding Training Framework for Seq2seq Models with Certified Robustness (2021.emnlp-main)
Copied to clipboard
| Challenge: | a recent study shows that inappropriate language can cause models to output profanity . authors propose a training framework to prevent such outputs from hurting the usability of models . |
| Approach: | proposed training framework eliminates the causes that trigger the generation of profanity . authors propose a framework that leverages a short list of profans to prevent this . |
| Outcome: | a proposed training framework can prevent models from generating profanity . the proposed framework leverages a short list of profanities examples . |