Papers by Chenyan Xiong
Distantly-Supervised Dense Retrieval Enables Open-Domain Question Answering without Evidence Annotation (2021.emnlp-main)
Copied to clipboard
| Challenge: | Open-domain question answering uses evidence retrieved from large corpus to answer questions . state-of-the-art approaches require intermediate evidence annotations for training . however, such intermediate annotations are expensive and methods that rely on them cannot transfer to the more common setting . |
| Approach: | They propose an open-domain question answering approach that alternately finds evidence from an up-to-date model and encourages the model to learn the most likely evidence. |
| Outcome: | The proposed approach improves over weak retrievers on multi-hop and single-hop benchmarks without using evidence labels. |
Less is More: Pretrain a Strong Siamese Encoder for Dense Text Retrieval Using a Weak Decoder (2021.emnlp-main)
Copied to clipboard
Shuqi Lu, Di He, Chenyan Xiong, Guolin Ke, Waleed Malik, Zhicheng Dou, Paul Bennett, Tie-Yan Liu, Arnold Overwijk
| Challenge: | Dense retrieval requires high-quality text sequence embeddings to support effective search in the representation space. |
| Approach: | They propose a self-learning method that pre-trains the autoencoder using a weak decoder to push the encoder to provide better sequence representations. |
| Outcome: | The proposed model significantly boosts the effectiveness and few-shot ability of dense retrieval models on web search, news recommendation, and open domain question answering. |
ResearchArena: Benchmarking Large Language Models’ Ability to Collect and Organize Information as Research Agents (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models excel across many natural language processing tasks but face challenges in domain-specific, analytical tasks such as conducting research surveys. |
| Approach: | They propose a benchmark to evaluate LLMs' capabilities in conducting research surveys. |
| Outcome: | The proposed benchmark is designed to evaluate LLMs' capabilities in conducting research surveys. |
Fine-grained Fact Verification with Kernel Graph Attention Network (2020.acl-main)
Copied to clipboard
| Challenge: | Existing methods for fact verification are based on dot-product attentions, but kernel-based attentions focus more on relevant evidence sentences and meaningful clues in the evidence graph. |
| Approach: | They propose a kernel-based attention network which conducts more fine-grained fact verification with kernel-basic attentions. |
| Outcome: | The proposed task achieves a 70.38% FEVER score and significantly outperforms existing fact verification models on FEVER, a large-scale benchmark for fact verification. |
COCO-DR: Combating Distribution Shift in Zero-Shot Dense Retrieval with Contrastive and Distributionally Robust Learning (2022.emnlp-main)
Copied to clipboard
| Challenge: | Using COCO-DR, we combat distribution shifts between source training tasks and target scenarios. |
| Approach: | They propose a method to combat distribution shifts between source training tasks and target scenarios by COtinuous COtrastive learning. |
| Outcome: | The proposed method outperforms existing models on BEIR and the giant GPT-3 embedding model with 500x more parameters. |
Efficient Multi-Agent System Training with Data Influence-Oriented Tree Search (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Model (LLM) based multi-agent systems (MAS) have high potential for tackling complex tasks through collaborative intelligence. |
| Approach: | They propose a framework that incorporates influence scores to guide tree search and data selection in data synthesis. |
| Outcome: | The proposed framework incorporates influence scores to guide tree search and data selection in data synthesis. |
Text Classification Using Label Names Only: A Language Model Self-Training Approach (2020.emnlp-main)
Copied to clipboard
| Challenge: | Current text classification methods require a large number of labeled documents as training data. |
| Approach: | They propose a model that uses only the label name of each class to train classification models on unlabeled data without using any labeled examples. |
| Outcome: | The proposed model achieves 90% accuracy on four benchmark datasets using label names as the only supervision . |
Adapting Open Domain Fact Extraction and Verification to COVID-FACT through In-Domain Language Modeling (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to verify scientifically false online information are limited by the lack of training data in the scientific domain. |
| Approach: | They propose an in-domain language modeling method for fact extraction and verification systems . they use SCIFACT to extract scientifically false online information . |
| Outcome: | The proposed method improves accuracy 30% on SCIFACT dataset . state-of-the-art model achieves only 46.6% precision, which is hard to be trusted for users. |
Dwell in the Beginning: How Language Models Embed Long Documents for Dense Retrieval (2024.acl-short)
Copied to clipboard
| Challenge: | Existing studies have shown that Transformer-based language models lose information in the middle of input sequences, especially in the context of web document retrieval. |
| Approach: | They examine position biases at multiple stages of the training pipeline for an encoder-decoder neural retrieval model, namely language model pre-training, contrastive pre- training, and contrastive fine-tuning. |
| Outcome: | The proposed model generates embeddings that better capture the beginning of the input content, with fine-tuning further aggravating this effect. |
Zero-Shot Dense Retrieval with Momentum Adversarial Domain Invariant Representations (2022.findings-acl)
Copied to clipboard
| Challenge: | Dense retrieval (DR) methods first encode texts into a dense embedding space and then conduct text retrieval using efficient nearest neighbor search. |
| Approach: | They propose Momentum adversarial Domain Invariant Representation learning to train a domain classifier that distinguishes source versus target domains and adversarially updates the DR encoder to learn domain invariant representations. |
| Outcome: | The proposed method outperforms baselines on 10+ ranking datasets collected in the BEIR benchmark in the zero-shot setting, with more than 10% relative gains on datasets with enough sensitivity for DR models’ evaluation. |
Target-Guided Open-Domain Conversation (P19-1)
Copied to clipboard
| Challenge: | a new study aims to improve opendomain chat systems by integrating goals and strategy into the system. |
| Approach: | They propose a structured approach that introduces coarse-grained keywords to control intended content of system responses and attains smooth conversation transition through turn-level supervised learning. |
| Outcome: | The proposed system produces meaningful and effective conversations significantly better than other approaches. |
Open Domain Web Keyphrase Extraction Beyond Language Modeling (D19-1)
Copied to clipboard
| Challenge: | Recent neural methods for keyphrase extraction are mostly observed in documents originating from the scientific domain. |
| Approach: | They develop a neural keyphrase extraction model that goes beyond language understanding to handle the variations of domain and content quality. |
| Outcome: | The proposed model can handle the variations of domain and content quality without restriction of the domain, quality, nor content of the documents. |
Multi-Step Reasoning Over Unstructured Text with Beam Dense Retrieval (2021.naacl-main)
Copied to clipboard
| Challenge: | Current methods for complex question answering use structured knowledge and unstructured text. |
| Approach: | They propose a multi-step retrieval approach that iteratively forms an evidence chain through beam search in dense representations. |
| Outcome: | The proposed method is competitive to state-of-the-art systems without using semi-structured information. |
Reduce Catastrophic Forgetting of Dense Retrieval Training with Teleportation Negatives (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent research shows strong empirical advantages of dense retrieval in various information access scenarios, such as openQA. |
| Approach: | They propose a method which accumulates momentum negatives from past iterations and approximates future iteration with lookahead negatives as "teleportations" on web search and OpenQA, ANCE-Tele outperforms previous state-of-the-art systems of similar size and eliminates the dependency on sparse retrieval negatives. |
| Outcome: | The proposed method outperforms previous state-of-the-art systems on web search and OpenQA and is competitive among systems with significantly more parameters. |
Augmenting Zero-Shot Dense Retrievers with Plug-in Mixture-of-Memories (2023.emnlp-main)
Copied to clipboard
| Challenge: | Using mixture-of-memory augmenting to augment language models improves model generalization but with diminishing return. |
| Approach: | They develop a mechanism that augments language models with mixture-of-memory Augmentation (MoMA) they augment strong T5-based retrievers with the option to "plug in" unseen memory at inference time. |
| Outcome: | The proposed model outperforms methods with larger model sizes on the BEIR benchmark and achieves comparable or even better performance than methods relying on target-specific pretraining. |
Structure-Aware Language Model Pretraining Improves Dense Retrieval on Structured Data (2023.findings-acl)
Copied to clipboard
| Challenge: | Structure Aware Dense Retrieval (SANTA) model encodes user queries and structured data in one universal embedding space for retrieving structured data. |
| Approach: | They propose to use structured data and unstructured data to encode queries and structured data in one universal embedding space for retrieving structured data. |
| Outcome: | The proposed model achieves state-of-the-art on code search and product search and conducts convincing results in the zero-shot setting. |
Craw4LLM: Efficient Web Crawling for LLM Pretraining (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing work discards over 90% of the raw data collected from web crawls, highlighting the inefficiency of current web crawlers in collecting LLM pretraining data. |
| Approach: | They propose a web crawling method that leverages the preference of LLMs as the priority score of the web crawler’s scheduler to obtain high-quality pretraining data. |
| Outcome: | The proposed method achieves high-quality pretraining data on a web graph containing 900 million webpages from a commercial search engine's index with just 21% URLs crawled. |
ThinkNote: Enhancing Knowledge Integration and Utilization of Large Language Models via Constructivist Cognition Modeling (2026.findings-eacl)
Copied to clipboard
Zhipeng Xu, Zhenghao Liu, Yukun Yan, Shuo Wang, Shi Yu, Zheni Zeng, Chaojun Xiao, Zhiyuan Liu, Ge Yu, Chenyan Xiong
| Challenge: | Large Language Models (LLMs) exhibit suboptimal behaviors and inconsistencies when exposed to unfamiliar external information, underscoring their limitations in effectively leveraging such knowledge. |
| Approach: | They propose a framework that enhances the external knowledge utilization of Large Language Models through a two-stage constructivist cognitive modeling process. |
| Outcome: | The proposed framework achieves a 10% improvement over baseline methods on various question-answering benchmarks. |
CompleQA: Benchmarking the Impacts of Knowledge Graph Completion Methods on Question Answering (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies have focused on Knowledge Graph Completion as an end in itself, neglecting its potential impact on subsequent applications. |
| Approach: | They propose a benchmark to assess the impact of representative KGC methods on Knowledge Graph Question Answering (KGQA) they use a knowledge graph with 3 million triplets across 5 distinct domains to evaluate their results. |
| Outcome: | The proposed benchmark compares four well-known methods with two state-of-the-art systems to assess the impact of incomplete knowledge graphs on KGQA. |
RAGViz: Diagnose and Visualize Retrieval-Augmented Generation (2024.emnlp-demo)
Copied to clipboard
| Challenge: | Large language models (LLMs) lack domain-specific knowledge and can cause hallucinations. |
| Approach: | They propose a RAG diagnosis tool that visualizes the attentiveness of the generated tokens in retrieved documents. |
| Outcome: | RAGViz provides token and document-level attention visualization and generation comparison upon context document addition and removal. |
Few-Shot Text Ranking with Meta Adapted Synthetic Weak Supervision (2021.acl-long)
Copied to clipboard
Si Sun, Yingzhuo Qian, Zhenghao Liu, Chenyan Xiong, Kaitao Zhang, Jie Bao, Zhiyuan Liu, Paul Bennett
| Challenge: | Neural information retrieval models have shown advanced results in many ranking scenarios where massive relevance labels or clickthrough data are available. |
| Approach: | They propose a domain adaptive learning method that generalizes Neu-IR models from label-rich source domains to few-shot target domains. |
| Outcome: | The proposed method improves the few-shot ranking accuracy of Neu-IR models on three TREC benchmarks in the web, news, and biomedical domains. |
Fusion-in-T5: Unifying Variant Signals for Simple and Effective Document Ranking with Attention Fusion (2024.lrec-main)
Copied to clipboard
| Challenge: | Current document ranking pipelines involve multiple ranking layers to integrate different information step-by-step. |
| Approach: | They propose a novel re-ranker Fusion-in-T5 which integrates text matching information, ranking features, and global document information into one single unified model via templated-based input and global attention. |
| Outcome: | The proposed model significantly improves ranking performance over complex cascade pipelines. |
Linking Knowledge to Care: Knowledge Graph-Augmented Medical Follow-Up Question Generation (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing large language models (LLMs) fail to identify information gaps across diverse symptoms. |
| Approach: | They propose a Knowledge Graph-augmented LLM with active in-context learning to generate relevant and important follow-up questions. |
| Outcome: | The proposed framework outperforms state-of-the-art methods by 5% - 8% on relevant benchmarks. |
Entity-Duet Neural Ranking: Understanding the Role of Knowledge Graph Semantics in Neural Information Retrieval (P18-1)
Copied to clipboard
| Challenge: | Entity-oriented search and neural-IR push the boundary of search engines from two different aspects. |
| Approach: | They propose an Entity-Duet Neural Ranking Model which integrates knowledge graphs into neural search systems. |
| Outcome: | The proposed model improves generalization ability of neural ranking models on a commercial search log. |
Long Document Ranking with Query-Directed Sparse Transformer (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to document ranking require long documents to be broken to fit in pretrained models. |
| Approach: | They propose a Query-Directed Sparse attention model that induces IR-axiomatic structures in transformer self-attention. |
| Outcome: | The proposed model enforces the principle properties desired in ranking while also enjoying efficiency from sparsity. |
Improving Multitask Retrieval by Promoting Task Specialization (2023.tacl-1)
Copied to clipboard
| Challenge: | despite its practical appeal, naive multitask retrieval lags behind task-specific retrieval. |
| Approach: | They propose to train a multitask retriever that promotes task specialization . the model is highly performant on the KILT benchmark . |
| Outcome: | The proposed model outperforms task-specific retrievals on the KILT benchmark . it learns parameters that are more task-specialized than naive retrieval without prompting or adaptive learning. |
Augmentation-Adapted Retriever Improves Generalization of Language Models as Generic Plug-In (2023.acl-long)
Copied to clipboard
| Challenge: | Prior work on retrieval augmentation fine-tuned the retriever and the LM, making them closely coupled. |
| Approach: | They propose a generic retrieval plug-in that can be used to fine-tune retrieval augmentation and a LM to learn a user's preferences. |
| Outcome: | The proposed retriever improves the generalization of large language models on the MMLU and PopQA datasets by learning LM’s preferences from a known source LM . |
Model-Generated Pretraining Signals Improves Zero-Shot Generalization of Text-to-Text Transformers (2023.acl-long)
Copied to clipboard
Linyuan Gong, Chenyan Xiong, Xiaodong Liu, Payal Bajaj, Yiqing Xie, Alvin Cheung, Jianfeng Gao, Xia Song
| Challenge: | Recent work in NLP has shown that pretrained language models have made notable progress toward generalization to unseen tasks. |
| Approach: | They propose to pretrain T5 using an auxiliary model to construct more challenging token replacements for the main model to denoise. |
| Outcome: | The proposed model outperforms similar-sized baseline models on prompted NLP benchmarks and rivals the state-of-the-art model with only **8%** of its parameters. |
On the Feasibility of In-Context Probing for Data Attribution (2025.findings-naacl)
Copied to clipboard
| Challenge: | In-context probing (ICP) can be used to identify training data that contributes to model outputs, but many data attribution methods, such as influence functions, use model gradients and are computationally expensive. |
| Approach: | They propose to use in-context probing (ICP) to proxy for gradient-based data attribution for data selection under conditions contingent on data similarity. |
| Outcome: | The proposed method can be used to identify training data that contribute to model outputs and fine tune models on training data. |
Fact-Aware Multimodal Retrieval Augmentation for Accurate Medical Radiology Report Generation (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing multimodal foundation models suffer from serious factual inaccuracy in radiology report generation. |
| Approach: | They propose a fact-aware multimodal retrieval-augmented pipeline for generating accurate radiology reports using RadGraph. |
| Outcome: | The proposed multimodal retrieval-augmented pipeline outperforms state-of-the-art retrievers on language generation and radiology-specific metrics. |
Dimension Reduction for Efficient Dense Retrieval via Conditional Autoencoder (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing work reserves the principle dimensions of query and document embeddings for building more efficient retrieval systems. |
| Approach: | They propose to use Conditional Autoencoder to compress high-dimensional embeddings to maintain the same embeddable distribution and better recover ranking features. |
| Outcome: | The proposed algorithm achieves comparable ranking performance with its teacher model and makes the retrieval system more efficient. |
TIAGE: A Benchmark for Topic-Shift Aware Dialog Modeling (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing dialog models can generate on-topic utterances but struggle to proactively switch topics. |
| Approach: | They propose a topic-shift aware dialog benchmark based on human topic shift annotations. |
| Outcome: | The proposed benchmark enables chatbots to generate topic-shift responses while still struggling to decide when to change topic. |
Automatic Event Salience Identification (D18-1)
Copied to clipboard
| Challenge: | Existing models for analyzing salience of discourse units are inadequate . authors propose two saliency detection models based on discourse relations . |
| Approach: | They propose two salience detection models based on discourse relations that capture complex interactions between discourse units. |
| Outcome: | The proposed models outperform the strong frequency baseline and improve the feature based model by a large margin. |
Grounded Conversation Generation as Guided Traverses in Commonsense Knowledge Graphs (2020.acl-main)
Copied to clipboard
| Challenge: | Existing models that generate natural language responses for conversations degenerate dull and repetitive contents, leading to off-topic and useless responses. |
| Approach: | They propose a conversation generation model which leverages commonsense knowledge graphs to explicitly model conversation flows by grounding conversations to the concept space. |
| Outcome: | Experiments on Reddit conversations show that the proposed model generates more semantic and informative responses while using 70% fewer parameters. |
Interpret and Control Dense Retrieval with Sparse Latent Features (2025.naacl-short)
Copied to clipboard
| Challenge: | Dense embeddings deliver strong retrieval performance but lack interpretability and controllability. |
| Approach: | They propose a novel approach using sparse autoencoders to interpret and control dense embeddings via latent sparsity. |
| Outcome: | The proposed approach retains the same retrieval accuracy as the original dense vectors, affirming their faithfulness. |
Toolink: Linking Toolkit Creation and Using through Chain-of-Solving on Open-Source Model (2024.naacl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have made significant progress in utilizing tools, but their closed-source nature and high inference costs pose limitations on their adaptability. |
| Approach: | They propose a framework that performs task-solving by first creating a toolkit and then integrating the planning and calling of tools through a chain-of-solve approach. |
| Outcome: | The proposed model performs task-solving by harnessing Toolink's creativity and CoS ability on ChatGPT and finetunes the LLaMA-7B model. |
Beneficial Reasoning Behaviors in Agentic Search and Effective Training Methods to Obtain Them (2026.findings-acl)
Copied to clipboard
| Challenge: | Agentic search requires large language models to perform multi-step searches to solve complex information needs. |
| Approach: | They propose a training approach that equips agentic search models with reasoning behaviors before reinforcement learning (RL) they compare successful and failed trajectories and propose supervised fine-tuning and standard RL . |
| Outcome: | The proposed approach outperforms direct RL by 37.2% on three web benchmarks and 6.2% on seven multi-hop QA benchmarks. |
Understand User Opinions of Large Language Models via LLM-Powered In-the-Moment User Experience Interviews (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing large language models (LLMs) are difficult to evaluate and often lack the ability to capture user opinions. |
| Approach: | They propose an LLM-powered interviewer that conducts in-the-moment user experience interviews right after users interact with LLMs and automatically gathers insights about user opinions from massive interview logs. |
| Outcome: | The proposed interviewer captures interesting user opinions, e.g., bipolar views on the displayed reasoning process of DeepSeek-R1 and demands for information freshness and multi-modality. |
Contrastive Multi-document Question Generation (2021.eacl-main)
Copied to clipboard
Woon Sang Cho, Yizhe Zhang, Sudha Rao, Asli Celikyilmaz, Chenyan Xiong, Jianfeng Gao, Mengdi Wang, Bill Dolan
| Challenge: | Multi-document question generation focuses on generating a question that covers the common aspect of multiple documents, but a naive model trained only using the targeted document set may generate too generic questions that cover a larger scope than delineated by the document set. |
| Approach: | They propose a contrastive learning strategy where given ‘positive’ and ‘negative’ sets of documents, generate a question that is closely related to the ‘positive' set but far away from the ‘negative' set. |
| Outcome: | The proposed model significantly outperforms several strong baselines, as measured by automatic metrics and human evaluation. |