Papers by Bang Liu
Resonance RoPE: Improving Context Length Generalization of Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have demonstrated their potential across a wide spectrum of natural language processing tasks. |
| Approach: | They propose a novel approach to narrow the generalization gap in TSTL scenarios by refining the interpolation of RoPE features for OOD positions. |
| Outcome: | The proposed approach improves performance without additional online computational costs on train-short-test-long scenarios. |
Enhancing Agent Learning through World Dynamics Modeling (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods assume that large language models have a complete understanding of their environment, overlooking potential gaps in their grasp of actual world dynamics. |
| Approach: | They propose a framework that discovers world dynamics from a small number of demonstrations, verifies the correctness of these dynamics, and evolves new, advanced dynamics tailored to the current situation. |
| Outcome: | The proposed framework discovers, verifies, and evolves world dynamics from a small number of demonstrations, and compares the automatically generated dynamics with human-annotated world dynamics. |
Data Interpreter: An LLM Agent for Data Science (2025.findings-acl)
Copied to clipboard
Sirui Hong, Yizhang Lin, Bang Liu, Bangbang Liu, Binhao Wu, Ceyao Zhang, Danyang Li, Jiaqi Chen, Jiayi Zhang, Jinlin Wang, Li Zhang, Lingyao Zhang, Min Yang, Mingchen Zhuge, Taicheng Guo, Tuo Zhou, Wei Tao, Robert Tang, Xiangtao Lu, Xiawu Zheng, Xinbing Liang, Yaying Fei, Yuheng Cheng, Yongxin Ni, Zhibin Gou, Zongze Xu, Yuyu Luo, Chenglin Wu
| Challenge: | Large Language Models (LLMs) excel in various domains but face challenges when applied to data science workflows due to their complex, multi-stage nature. |
| Approach: | They propose a hierarchical graph-based agent that represents complexity and a progressive strategy for step-by-step verification, refinement, and consistent context management. |
| Outcome: | The proposed agent surpasses state-of-the-art baselines on the MATH dataset and performs better on InfiAgent-DABench. |
Semantic and Syntactic Enhanced Aspect Sentiment Triplet Extraction (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to extract triplets from sentences neglect the mutual information between aspects and have the problem of error propagation. |
| Approach: | They propose a Semantic and Syntactic Enhanced aspect Sentiment triplet Extraction model to exploit the syntactical and semantic relationships between the triplet elements and jointly extract them. |
| Outcome: | The proposed model outperforms existing methods on four benchmark datasets and significantly outperformed existing approaches. |
Enhancing Healthcare LLM Trust with Atypical Presentations Recalibration (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for eliciting and calibrating large language models have focused on general reasoning datasets, yielding only modest improvements. |
| Approach: | They propose a method which leverages atypical presentations to adjust model confidence estimates. |
| Outcome: | The proposed method reduces calibration errors by approximately 60% on three medical question answering datasets and outperforms existing methods such as vanilla verbalized confidence, CoT verbalised confidence and others. |
Explore Spurious Correlations at the Concept Level in Language Models for Text Classification (2024.acl-long)
Copied to clipboard
| Challenge: | Language models have demonstrated remarkable performance in numerous NLP tasks, employing both fine-tuning and in-context learning (ICL) methods. |
| Approach: | They propose a method to assess concept bias in models during fine-tuning and in-context learning using ChatGPT. |
| Outcome: | The proposed method outperforms token removal approaches and is validated through extensive testing. |
Guiding the Growth: Difficulty-Controllable Question Generation through Step-by-Step Rewriting (2021.acl-long)
Copied to clipboard
| Challenge: | Existing QG systems perform substantially worse in answering multi-hop questions than single-hop ones. |
| Approach: | They propose a framework that progressively increases question difficulty through step-by-step rewriting under the guidance of an extracted reasoning chain. |
| Outcome: | The proposed framework increases question difficulty through step-by-step rewriting under the guidance of an extracted reasoning chain. |
Integrating Semantics and Neighborhood Information with Graph-Driven Generative Models for Document Retrieval (2021.acl-long)
Copied to clipboard
Zijing Ou, Qinliang Su, Jianxing Yu, Bang Liu, Jingwen Wang, Ruihui Zhao, Changyou Chen, Yefeng Zheng
| Challenge: | Existing methods for document hashing combine only one of semantics and neighborhood information, lacking a theoretical principle to guide the integration process. |
| Approach: | They propose to encode neighborhood information with a graph-induced Gaussian distribution and integrate it with generative models. |
| Outcome: | The proposed model can be trained as efficiently as state-of-the-art methods on benchmark datasets. |
Deciphering Digital Detectives: Understanding LLM Behaviors and Capabilities in Multi-Agent Mystery Games (2024.findings-acl)
Copied to clipboard
| Challenge: | In this study, we explore the application of Large Language Models (LLMs) in Jubensha, a Chinese detective role-playing game and a novel area in Artificial Intelligence (AI) driven gaming. |
| Approach: | They propose to use large language models to foster AI agent development in Jubensha, a Chinese detective role-playing game. |
| Outcome: | The proposed framework enables AI agents to engage in Jubensha games autonomously. |
Efficient Classification of Long Documents via State-Space Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing transformer-based models can only process long documents with limited computational resources due to their quadratic computation time and space. |
| Approach: | They propose to use state-space models for long document classification tasks instead of using sparse or hierarchical structures to solve this problem. |
| Outcome: | The proposed model performs comparable to self-attention models while being 36% more efficient. |
SkillQG: Learning to Generate Question for Reading Comprehension Assessment (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing question generation systems focus on the literal nature of questions and rarely consider comprehension types of the generated questions. |
| Approach: | They propose a question generation framework with controllable comprehension types for machine reading comprehension models. |
| Outcome: | Empirical results show that SkillQG outperforms baselines in quality, relevance, and skill-controllability while showing a performance boost in downstream question answering task. |
Imperfect also Deserves Reward: Multi-Level and Sequential Reward Modeling for Better Dialog Management (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing research on taskoriented dialog systems mainly includes pipeline and end-to-end methods due to its non-differentiable nature. |
| Approach: | They propose a multi-level reward modeling approach that factorizes a reward into a three-level hierarchy: domain, act, and slot. |
| Outcome: | The proposed approach significantly improves performance and speed of training in a wide range of dialog systems. |
S2R: Teaching LLMs to Self-verify and Self-correct via Reinforcement Learning (2025.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to incentivize LLMs’ deep thinking abilities require large-scale data or significant training efforts. |
| Approach: | They introduce an efficient framework that enhances LLM reasoning by teaching models to self-verify and self-correct during inference. |
| Outcome: | The proposed framework outperforms models trained on long-CoT distilled data with 3.1k initialization samples and achieves an accuracy improvement of 51.0% to 81.6%. |
Fine-tuning Happens in Tiny Subspaces: Exploring Intrinsic Task-specific Subspaces of Pre-trained Language Models (2023.acl-long)
Copied to clipboard
| Challenge: | Pre-trained language models are overly parameterized and have significant redundancy . recent studies show that PLMs are highly over-parameterized and robust to pruning . |
| Approach: | They propose to re-parameter and fine-tune pre-trained language models from a new perspective: Discovery of intrinsic task-specific subspace. |
| Outcome: | The proposed model can be fine-tuned in the subspace with a small number of free parameters. |
Improving Multi-turn Emotional Support Dialogue Generation with Lookahead Strategy Planning (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing research on building ES conversation systems only considered single-turn interactions with users, which is over-simplified and has limited support for multi-turn systems. |
| Approach: | They propose a multi-turn ES conversation system that uses lookahead heuristics to estimate future user feedback after using particular strategies. |
| Outcome: | The proposed system significantly outperforms baselines in both dialogue generation and strategy planning. |
Multimodal Prompt Learning for Product Title Generation with Extremely Limited Labels (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to generate informative titles for products with limited labels are inadequate for novel products. |
| Approach: | They propose a prompt-based approach to generate attractive titles for novel products . they use multimodal prompts to preserve characteristics and writing styles of novel products. |
| Outcome: | The proposed approach achieves state-of-the-art results on novel product categories with limited labels. |
MEDSYN: Benchmarking Multi-EviDence SYNthesis in Complex Clinical Cases for Multimodal Large Language Models (2026.findings-acl)
Copied to clipboard
Boqi Chen, Xudong Liu, Jiachuan Peng, Marianne Frey-Marti, Kyle Lam, Bang Zheng, Lin Li, Jianing Qiu
| Challenge: | Existing benchmarks for multimodal large language models do not capture real-world clinical complexity. |
| Approach: | They evaluate multilingual, multimodal multimodal models of clinical cases with up to 7 distinct visual clinical evidence types per case. |
| Outcome: | The proposed model outperforms human models on differential diagnosis (DDx) generation and final diagnosis (FDx) selection. |
INDOORWORLD : Integrating Physical Task Solving and Social Simulation in A Heterogeneous Multi-Agent Environment (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing virtual environments for LLM agent research focus on task solving or social simulation . existing environments for virtual environments lack physical grounding of social behaviors . |
| Approach: | They propose a virtual environment that tightly integrates physical and social dynamics . IndoorWorld is a heterogeneous multi-agent environment that integrates social and physical dynamics based on a simulation of physical environments . |
| Outcome: | The proposed environment integrates physical and social dynamics into a heterogeneous multi-agent environment. |
R3Mem: Bridging Memory Retention and Retrieval via Reversible Compression (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing memory solutions that store information via parameters struggle with reliable retrieval. |
| Approach: | They propose a memory network that optimizes both information Retention and Retrieval through Reversible context compression. |
| Outcome: | The proposed memory network outperforms conventional memory modules in long-horizon interaction tasks like conversational agents and achieves state-of-the-art performance in language modeling and retrieval-augmented generation tasks. |
FACT: Examining the Effectiveness of Iterative Context Rewriting for Multi-fact Retrieval (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are proficient at retrieving single facts from extended contexts, but struggle with tasks requiring simultaneous retrieval of multiple facts. |
| Approach: | They propose a method that refines context through successive rounds of rewriting to address this problem by finding all Crucial Texts (FACT) |
| Outcome: | The proposed method improves multi-fact retrieval performance across tasks, though improvements are less notable in general-purpose QA scenarios. |
MatSci-NLP: Evaluating Scientific Language Models on Materials Science Language Tasks Using Text-to-Schema Modeling (2023.acl-long)
Copied to clipboard
| Challenge: | Using publicly available materials science text data, we construct a benchmark for evaluating the performance of natural language processing (NLP) models on materials science texts. |
| Approach: | They propose a natural language benchmark for evaluating the performance of natural language processing (NLP) models on materials science text. |
| Outcome: | The proposed model outperforms BERT-based models on scientific text and a model pretrained on materials science journals. |
Refining BERT Embeddings for Document Hashing via Mutual Information Maximization (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing unsupervised document hashing methods are mostly established on generative models . due to the difficulties of capturing long dependency structures, these methods rarely model the raw documents directly . |
| Approach: | They propose to learn hash codes from BERT embeddings by modifying existing models . they use mutual information maximization principle to maximize mutual information . |
| Outcome: | The proposed method outperforms existing methods learned from BERT embeddings on three benchmark datasets. |
MultiCapCLIP: Auto-Encoding Prompts for Zero-Shot Multilingual Visual Captioning (2023.acl-long)
Copied to clipboard
| Challenge: | Existing methods for supervised visual captioning require large scale of images or videos paired with descriptions in a specific language. |
| Approach: | They propose a zero-shot approach that generates captions for different scenarios without labeling . they use concept prompts to retrieve concepts and auto-encode them to learn writing styles . |
| Outcome: | The proposed approach generates captions for different scenarios and languages without labeled vision-caption pairs. |
MAPO: Boosting Large Language Model Performance with Model-Adaptive Prompt Optimization (2023.findings-emnlp)
Copied to clipboard
Yuyan Chen, Zhihao Wen, Ge Fan, Zhengyu Chen, Wei Wu, Dayiheng Liu, Zhixu Li, Bang Liu, Yanghua Xiao
| Challenge: | Existing research emphasizes the importance of adapting prompts to specific tasks, rather than specific LLMs. |
| Approach: | They propose a model-adaptive prompt optimizer method that optimizes original prompts for each LLM in downstream tasks. |
| Outcome: | The proposed method can optimize prompts for an LLM in downstream tasks. |
Mind’s Mirror: Distilling Self-Evaluation Capability and Comprehensive Thinking from Large Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have achieved significant advances in natural language processing, but their scale and computational demands pose challenges to their practical application. |
| Approach: | They propose a method for distilling the self-evaluation capability from LLMs into SLMs and advocate for more comprehensive thinking by incorporating multiple distinct CoTs and self-estimation outputs. |
| Outcome: | The proposed method significantly improves the performance of distilled SLMs on three NLP benchmarks. |
Search-Oriented Conversational Query Editing (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing CQR models are not learned toward improving the downstream search performance . existing models generate the rewrite token-by-token from scratch . |
| Approach: | They propose a text editing-based CQR model tailored for conversational search . they propose rewrite tokens are selected from the dialogue in a non-autoregressive fashion . |
| Outcome: | The proposed model outperforms state-of-the-art models on three conversational search benchmarks while having low rewriting latency. |
ICA-Proto: Iterative Cross Alignment Prototypical Network for Incremental Few-Shot Relation Classification (2023.findings-eacl)
Copied to clipboard
Wangjie Jiang, Zhihao Ye, Bang Liu, Ruihui Zhao, Jianguang Zheng, Mengyao Li, Zhiyong Li, Yujiu Yang, Yefeng Zheng
| Challenge: | Existing methods for relation classification suffer from the scarcity of manually annotated data. |
| Approach: | They propose a novel relation classification model that incorporates query representation into the encoding of novel prototypes and utilizes iteratively to achieve more interaction. |
| Outcome: | The proposed model outperforms the state-of-the-art model on two benchmark datasets. |
Encoding Hierarchical Schema via Concept Flow for Multifaceted Ideology Detection (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on ideology detection focus on one generic facet and ignore label semantics and explanatory descriptions of ideologies. |
| Approach: | They propose a concept semantics-enhanced framework for multifaceted ideology detection . it enables concepts to flow across levels of the schema tree and enriches concept representations with multi-granularity semantics. |
| Outcome: | The proposed framework achieves state-of-the-art in the cross-topic scenario and on the benchmark dataset. |
AceGPT, Localizing Large Language Models in Arabic (2024.naacl-long)
Copied to clipboard
Huang Huang, Fei Yu, Jianqing Zhu, Xuening Sun, Hao Cheng, Song Dingjie, Zhihong Chen, Mosen Alharthi, Bang An, Juncai He, Ziche Liu, Junying Chen, Jianquan Li, Benyou Wang, Lian Zhang, Ruoyu Sun, Xiang Wan, Haizhou Li, Jinchao Xu
| Challenge: | Significant concerns emerge when addressing cultural sensitivity and local values. |
| Approach: | They propose a localized Large Language Model (LLM) specifically for Arabic, a language imbued with unique cultural characteristics inadequately addressed by current mainstream models. |
| Outcome: | The proposed model sets the state-of-the-art standard for open Arabic LLMs across various benchmarks. |
LLMs for Bayesian Optimization in Scientific Domains: Are We There Yet? (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models have been proposed as general-purpose agents for experimental design . eval: LLMs show no sensitivity to experimental feedback. |
| Approach: | They propose a method that combines LLM prior knowledge with nearest-neighbor sampling to guide the design of experiments. |
| Outcome: | The proposed method outperforms classical methods in the design of experiments. |
Matching Article Pairs with Graphical Decomposition and Convolutions (P19-1)
Copied to clipboard
| Challenge: | Existing methods for matching sentence pairs do not perform well in longer documents . Existing approaches for matching sentences do not work in longer document understanding tasks . |
| Approach: | They propose to model article pairs by comparing sentences that enclose same concept vertex . they propose to use a concept interaction graph to match articles by encoding sentences . |
| Outcome: | The proposed methods show significant improvements over existing methods . the proposed datasets consist of 30K pairs of breaking news articles . |
FAC2E: Better Understanding Large Language Model Capabilities by Dissociating Language and Cognition (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) are evaluated by overall performance on various text understanding and generation tasks. |
| Approach: | They propose a framework for Fine-grAined and Cognition-grounded LLMs’ Capability Evaluation that dissociates the language-related capabilities from cognition-related ones. |
| Outcome: | The proposed framework dissociates the language-related capabilities from cognition-related ones and breaks down the process of applying a specific capability into three sub-steps: recalling relevant knowledge, utilizing knowledge, and solving problems. |
O2NA: An Object-Oriented Non-Autoregressive Approach for Controllable Video Captioning (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for video captioning consider a sequence of frames and biases towards focused objects. |
| Approach: | They propose an Object-Oriented Non-Autoregressive approach to video captioning . it performs three steps: 1) identify the focused objects and predict their locations . 2) generate related attribute words and relation words of these focused objects to form a draft caption . |
| Outcome: | The proposed method achieves competitive results with the state-of-the-art methods but with higher diversity and faster inference speed. |
HoneyComb: A Flexible LLM-Based Agent System for Materials Science (2024.findings-emnlp)
Copied to clipboard
| Challenge: | specialized large language models (LLMs) have shown promise in materials science but often struggle with the distinct complexities of materials science tasks. |
| Approach: | They propose a new LLM-based agent system specifically designed for materials science that leverages a reliable materials science knowledge base and a sophisticated tool hub. |
| Outcome: | The proposed system outperforms baseline models across tasks in materials science while ensuring accuracy and relevance. |
Feeding What You Need by Understanding What You Learned (2022.acl-long)
Copied to clipboard
| Challenge: | Existing research on machine reading comprehension rely heavily on large-size models and corpus to improve performance. |
| Approach: | They propose a framework that assesses model capabilities in an explainable and multi-dimensional manner. |
| Outcome: | The proposed framework achieves an 11.22% / 8.71% improvement of EM / F1 on MRC tasks. |
Seeing Beyond Words: MatVQA for Challenging Visual-Scientific Reasoning in Materials Science (2026.findings-acl)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) outperform existing benchmarks in both natural language and coding domains. |
| Approach: | They propose a scalable benchmark that integrates vision and language modalities to address this gap by eliminating textual shortcuts. |
| Outcome: | The new benchmark outperforms existing benchmarks in both natural language and coding domains. |
Improving Context Fidelity via Native Retrieval-Augmented Reasoning (2025.emnlp-main)
Copied to clipboard
Suyuchen Wang, Jinlin Wang, Xinyu Wang, Shiqi Li, Xiangru Tang, Sirui Hong, Xiao-Wen Chang, Chenglin Wu, Bang Liu
| Challenge: | Existing approaches to fidelity to contexts rely on expensive supervised fine-tuning to generate evidence post-answer or train models to perform web searches without improving utilization of the given context. |
| Approach: | They propose a native retrieval-augmented reasoning framework that integrates in-context evidence with the model’s own retrieval capabilities. |
| Outcome: | The proposed approach outperforms supervised fine-tuning, retrieval-augmented generation methods, and external retrieval solutions on multiple real-world and counterfactual QA benchmarks. |
Ideology Takes Multiple Looks: A High-Quality Dataset for Multifaceted Ideology Detection (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing datasets for the ID task only label a text as ideologically left- or right-leaning as a whole, regardless whether the text containing one or more different issues. |
| Approach: | They construct an ideological schema for a multifaceted ideology detection task using MITweet and an English Twitter dataset. |
| Outcome: | The proposed task uses a MITweet dataset with 12,594 English Twitter posts, each annotated with a Relevance and an Ideology label for all twelve facets. |
Identifying while Learning for Document Event Causality Identification (2024.acl-long)
Copied to clipboard
| Challenge: | Existing studies focus on causality existence, but ignore causal direction. |
| Approach: | They propose a new *identifying while learning* mode for the ECI task that takes care of the causal direction and updates events’ representations for boosting next round of causality identification. |
| Outcome: | The proposed method outperforms the state-of-the-art methods on two public datasets. |
OPEx: A Component-Wise Analysis of LLM-Centric Agents in Embodied Instruction Following (2024.acl-long)
Copied to clipboard
| Challenge: | Embodied Instruction Following (EIF) is a crucial task in embodied learning . however, there is n'a unified understanding regarding the impact of various components on task performance . |
| Approach: | They propose a framework that delineates the core components essential for embodied learning tasks . they integrate a multi-agent design into the Planner component of their LLM-centric architecture . |
| Outcome: | OPEx delineates the core components essential for solving embodied learning tasks . integrating a multi-agent design into the Planner component of the LLM-centric architecture further elevates performance. |
HoneyBee: Progressive Instruction Finetuning of Large Language Models for Materials Science (2023.findings-emnlp)
Copied to clipboard
| Challenge: | LLaMa-based language model for materials science is first of its kind in the world . |
| Approach: | They propose an instruction-based process for trustworthy data curation in materials science (MatSci-Instruct) they then apply this process to finetune a LLaMa-based language model targeted for materials science. |
| Outcome: | The proposed model outperforms existing language models on materials science tasks and improves in successive stages of refinement. |
Concise Math Reasoning via Difficulty-Aware Distillation (2026.findings-acl)
Copied to clipboard
Yifan Wu, Jingze Shi, Bingheng Wu, Jiayi Zhang, Xiaotian Lin, Yizhang Zhu, Zhaoyang Yu, Bang Liu, Chenglin Wu, Nan Tang, Yuyu Luo
| Challenge: | Human experts tackle difficult math problems by identifying and executing a few pivotal steps rather than listing every intermediate thought. |
| Approach: | They propose a method for producing training data that mirrors concise human reasoning by rewriting a problem's solution to retain only the essential steps. |
| Outcome: | The proposed method outperforms models trained on 800k long CoT and cuts training and inference costs. |
QRelScore: Better Evaluating Generated Questions with Deeper Understanding of Context-aware Relevance (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing metrics for assessing question generation fail to take into account the input context of generation. |
| Approach: | They propose a context-aware Relevance evaluation metric for Question Generation that takes into account the context of question generation into account. |
| Outcome: | The proposed metric achieves higher correlation with human judgments while being much more robust to adversarial samples. |