Papers by Wenhan Wang
HybridQA: A Dataset of Multi-Hop Question Answering over Tabular and Textual Data (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing question answering datasets focus on dealing with homogeneous information, but using homogenous information alone might lead to coverage problems. |
| Approach: | They propose a large-scale question-answering dataset that requires reasoning on heterogeneous information. |
| Outcome: | The proposed model can achieve an EM score of 40% while the existing model is far behind human performance. |
CoRanking: Collaborative Ranking with Small and Large Ranking Agents (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Listwise ranking based on Large Language Models (LLMs) has achieved state-of-the-art performance in Information Retrieval (IR) however, their effectiveness often depends on LLMs with massive parameter scales and computationally expensive sliding window processing, leading to substantial efficiency bottlenecks. |
| Approach: | They propose a Collaborative Ranking framework (CoRanking) for LLM-based listwise ranking based on large language models with massive parameter scales and computationally expensive sliding window processing. |
| Outcome: | The proposed framework reduces ranking latency by approximately 70% while improving effectiveness compared to the standalone large reranker. |
Self-Supervised Learning for Contextualized Extractive Summarization (P19-1)
Copied to clipboard
| Challenge: | Existing models for extractive summarization are usually trained from scratch with a cross-entropy loss . previous work builds an end-to-end system to learn to choose sentences without explicitly modeling document context . |
| Approach: | They propose three auxiliary pre-training tasks that learn to capture the document context in a self-supervised fashion. |
| Outcome: | The proposed models outperform existing models on a CNN/DM dataset. |
Simple yet Effective Bridge Reasoning for Open-Domain Multi-Hop Question Answering (D19-58)
Copied to clipboard
| Challenge: | Existing work on open-domain multi-hop question answering relies on off-the-shelf information retrieval techniques to retrieve answer passages. |
| Approach: | They propose a new subproblem for open-domain multi-hop question answering . they aim to recognize the anchor from a set of start passages with a reading comprehension model . |
| Outcome: | The proposed method significantly improves the baseline method on the open-domain hotpotQA benchmark. |
LM-Infinite: Zero-Shot Extreme Length Generalization for Large Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Currently, large language models (LLMs) train on short text segments due to the computational overhead quadratic in the input lengths of their Transformer architectures. |
| Approach: | They propose a method that allows LLMs pre-trained with 2K or 4K-long segments to generalize to up to 200M length inputs while retaining perplexity. |
| Outcome: | The proposed method achieves 2.7 decoding speed up and 7.5 memory saving over the original model. |
Do Multi-hop Readers Dream of Reasoning Chains? (D19-58)
Copied to clipboard
| Challenge: | Existing models for multihop reasoning are limited in their performance . multi-hop reasoning requires the ability to gather information from multiple passages . |
| Approach: | They propose a method that provides the full reasoning chain of multiple passages instead of just one final passage where the answer appears. |
| Outcome: | The proposed model improves on existing models by providing the full reasoning chain of multiple passages instead of just one final passage where the answer appears. |
Coarse-to-Fine Contrastive Learning in Image-Text-Graph Space for Improved Vision-Language Compositionality (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have highlighted severe limitations of contrastive learning models in their ability to perform compositional reasoning over objects, attributes, and relations. |
| Approach: | They propose a graph decomposition framework and negative mining techniques to improve attribute binding and relation understanding of scene graphs. |
| Outcome: | The proposed approach improves attribute binding, relation understanding, generalization, and productivity on multiple benchmarks. |
AGSC: Adaptive Granularity and Semantic Clustering for Uncertainty Quantification in Long-text Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for aggregating large-form outputs overlook the nuance of neutral information and suffer from the high computational cost of fine-grained decomposition. |
| Approach: | They propose a UQ framework that uses NLI neutral probabilities as triggers to distinguish irrelevance from uncertainty, reducing computation costs. |
| Outcome: | Experiments on BIO and LongFact show that the proposed framework reduces inference time by 60% compared to full atomic decomposition. |
One-Shot Relational Learning for Knowledge Graphs (D18-1)
Copied to clipboard
| Challenge: | Existing studies on knowledge graph completion require a large number of positive examples for each relation, but long-tail relations are more common in KGs and those newly added relations do not have many known triples for training. |
| Approach: | They propose a one-shot relational learning framework that utilizes the knowledge distilled by embedding models and learns a matching metric by considering both the learned embeddments and one-hop graph structures. |
| Outcome: | The proposed framework improves on existing embedding models and eliminates the need for retraining when dealing with newly added relations. |
Learning to Learn and Predict: A Meta-Learning Approach for Multi-Label Classification (D19-1)
Copied to clipboard
| Challenge: | Existing models for multi-label classification ignore complexity and dependencies among labels . Experimental results show that our method can obtain more accurate multi-lab classification results. |
| Approach: | They propose a meta-learning method to capture complex label dependencies . they use a Meta-learner to jointly learn the training policies and prediction policies for different labels. |
| Outcome: | The proposed method can capture complex label dependencies on fine-grained entity typing and text classification tasks. |
Towards Hierarchical Multi-Step Reward Models for Enhanced Reasoning in Large Language Models (2026.findings-acl)
Copied to clipboard
Teng Wang, Jiang Zhangyi, Zhenqi He, Hailei Gong, Shenyang Tong, Wenhan Yang, Zeyu Li, Yanan Zheng, Zifan He, Zewen Ye, Shengjie Ma, Jianping Zhang
| Challenge: | Existing Process Reward Models (PRMs) are vulnerable to reward hacking and require expensive, large-scale annotation of reasoning steps. |
| Approach: | They propose a reward model approach which evaluates both individual and consecutive reasoning steps from fine-grained and coarse-grounded level. |
| Outcome: | Empirical results show that the proposed model performs better than existing PRMs and is more robust than existing models. |
Evaluating the Validity of Word-level Adversarial Attacks with Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing adversarial examples can generate invalid adversarials due to significant changes in semantic meanings compared to their originals. |
| Approach: | They propose to use a large language model to evaluate adversarial examples by semantic constraints. |
| Outcome: | The proposed method can generate valid adversarial examples even when they are not equipped with semantic constraints. |
Sentence Embedding Alignment for Lifelong Relation Extraction (N19-1)
Copied to clipboard
| Challenge: | Existing approaches to relation extraction require a fixed set of relations . Existing methods assume a closed set of relationships and perform once-and-for-all training on a set of datasets. |
| Approach: | They propose to improve the stochastic gradient methods with a replay memory to alleviate the forgetting problem by anchoring the sentence embedding space. |
| Outcome: | The proposed method outperforms state-of-the-art methods on multiple benchmarks. |
Improving Question Answering over Incomplete KBs with Knowledge-Aware Reader (P19-1)
Copied to clipboard
| Challenge: | Existing models that use incomplete knowledge bases and text data to answer open-domain questions are insufficient to cover full evidence. |
| Approach: | They propose a model which learns to aggregate answer evidence from incomplete knowledge bases and text snippets. |
| Outcome: | The proposed model improves on the widely-used KBQA benchmark WebQSP across settings with different extents of incompleteness. |
Zero-shot Fact Verification by Claim Generation (2021.acl-short)
Copied to clipboard
| Challenge: | Existing methods for fact verification require large datasets, which can be expensive. |
| Approach: | They propose a framework for training a robust fact verification model by using automatically generated claims that can be supported, refuted, or unverifiable from evidence from Wikipedia. |
| Outcome: | The proposed framework reduces the demand for human-annotated training data and improves a model's F1 from 50% to 77%, equivalent in performance to 2K+ manually-curated examples. |
TESTEVAL: Benchmarking Large Language Models for Test Case Generation (2025.findings-naacl)
Copied to clipboard
Wenhan Wang, Chenyuan Yang, Zhijie Wang, Yuheng Huang, Zhaoyang Chu, Da Song, Lingming Zhang, An Ran Chen, Lei Ma
| Challenge: | Existing methods to generate test cases using large language models are limited in their ability to generate unit test cases. |
| Approach: | They propose a test case generation benchmark that uses large language models to generate unit test cases. |
| Outcome: | The proposed test case generation benchmarks compare LLMs with commercial and open-source LLM platforms and find that they lack the ability to comprehend program logic and execution paths. |
Unsupervised Multi-hop Question Answering by Question Generation (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing training data for multi-hop question answering (QA) is time-consuming and resource-intensive. |
| Approach: | They propose an unsupervised framework that generates human-like multi-hop training data from homogeneous and heterogeneously data sources. |
| Outcome: | The proposed framework achieves 61% and 83% of the supervised learning performance for the HybridQA and HotpotQA datasets. |
HSGraphAgent: Knowledge-Graph-Guided Large Language Models for Harmonized System Code Classification (2026.acl-long)
Copied to clipboard
| Challenge: | Harmonized System (HS) code classification is a hierarchically structured and regulation-constrained task, often complicated by short and noisy product descriptions. |
| Approach: | They propose a knowledge-graph-guided LLM framework that formulates HS classification as a stepwise, regulation-aware reasoning process over an explicit HS knowledge graph. |
| Outcome: | The proposed framework constrains inference to legally valid paths while producing explicit and traceable reasoning trajectories. |
Bridging the Training-Inference Gap for Dense Phrase Retrieval (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for building dense retrievers are often misaligned and do not reflect retrieval scenario at inference time. |
| Approach: | They propose a way to validate dense retrievers using a small subset of the entire corpus. |
| Outcome: | The proposed model improves top-1 phrase retrieval accuracy by 2 3 points and top-20 passage retrieval by 2 4 points for open-domain question answering. |
Sliding Windows Are Not the End: Exploring Full Ranking with Long-Context Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for listwise passage ranking use sliding window approach, which is inefficient as it requires repetitive and serialized processing. |
| Approach: | They propose a listwise label construction approach and importance-aware learning objective for full ranking. |
| Outcome: | The proposed method outperforms existing methods in listwise ranking tasks. |
Progressively Pretrained Dense Corpus Index for Open-Domain Question Answering (2021.eacl-main)
Copied to clipboard
| Challenge: | Existing open-domain question answering systems are insufficient to capture deep semantic matching that goes beyond lexical overlaps. |
| Approach: | They propose a sample-efficient method to pretrain the paragraph encoder using an existing pretraining model instead of heuristically created pseudo question-paragraph pairs. |
| Outcome: | The proposed method outperforms a strong dense retrieval baseline that uses 6 times more computation for training. |
Harnessing Pre-Trained Neural Networks with Rules for Formality Style Transfer (D19-1)
Copied to clipboard
| Challenge: | Existing studies normalize informal sentences with rules, but they introduce noise if we use them in a naive way. |
| Approach: | They propose to harness rules into a state-of-the-art neural network that is typically pretrained on massive corpora. |
| Outcome: | The proposed method can be used to generate a state-of-the-art on a small dataset. |
Effective Long-Context Scaling of Foundation Models (2024.naacl-long)
Copied to clipboard
Wenhan Xiong, Jingyu Liu, Igor Molybog, Hejia Zhang, Prajjwal Bhargava, Rui Hou, Louis Martin, Rashi Rungta, Karthik Abinav Sankararaman, Barlas Oguz, Madian Khabsa, Han Fang, Yashar Mehdad, Sharan Narang, Kshitiz Malik, Angela Fan, Shruti Bhosale, Sergey Edunov, Mike Lewis, Sinong Wang, Hao Ma
| Challenge: | Large language models (LLMs) are rapidly deployed and continue to evolve through scaling. |
| Approach: | They propose a method to train strong long-context LLMs that are capable of utilizing massive context windows of up to 32,000 tokens. |
| Outcome: | The proposed model can surpass gpt-3.5-turbo-16k's overall performance on long-context benchmarks with a cost-effective instruction tuning procedure that is free of expensive annotations. |
Imposing Label-Relational Inductive Bias for Extremely Fine-Grained Entity Typing (N19-1)
Copied to clipboard
| Challenge: | Existing entity typing systems exploit type hierarchy provided by KB schema to model label correlations. |
| Approach: | They propose a graph layer that encodes global label co-occurrence statistics and word-level similarities. |
| Outcome: | The proposed model achieves a 15.3% relative F1 improvement on a large dataset with over 10,000 free-form types. |
Relevant or Random: Can LLMs Truly Perform Analogical Reasoning? (2025.findings-acl)
Copied to clipboard
| Challenge: | Analogical reasoning is a unique ability of humans to address unfamiliar challenges by transferring strategies from relevant past experiences. |
| Approach: | They propose to use self-generated random examples to improve performance on a variety of reasoning tasks by incorporating relevant examples from relevant past experiences. |
| Outcome: | The proposed methods achieve comparable or even better performance on GSM8K with random biological examples. |
TWEETQA: A Social Media Focused Question Answering Dataset (P19-1)
Copied to clipboard
Wenhan Xiong, Jiawei Wu, Hong Wang, Vivek Kulkarni, Mo Yu, Shiyu Chang, Xiaoxiao Guo, William Yang Wang
| Challenge: | Social media is becoming an important realtime information source, especially during natural disasters and emergencies. |
| Approach: | They present a large-scale dataset for question answering over social media data . they gather tweets used by journalists and ask human annotators to write questions upon them . |
| Outcome: | The proposed dataset shows that neural models that perform well on formal texts are limited in their performance . the proposed model is still lagging behind human performance with a large margin . |
Text-guided 3D Human Generation from 2D Collections (2023.findings-emnlp)
Copied to clipboard
| Challenge: | 3D human modeling is used for engaging interaction in gaming, film, and animation. however, the customization of characters is crucial for creativity and scalability. |
| Approach: | They propose a 3D human generation using fashion descriptions to enhance 3D geometry transformation and fine-grained consistency. |
| Outcome: | The proposed model can generate a 3D human, guided by a fashion description, with high efficiency. |
VideoVista-CulturalLingo: 360° Horizons-Bridging Cultures, Languages, and Domains in Video Comprehension (2025.acl-long)
Copied to clipboard
| Challenge: | Existing video evaluation benchmarks focus on a single language, typically English, and feature videos rooted in Western cultural contexts. |
| Approach: | They propose a video evaluation benchmark designed to bridge cultural, linguistic, and domain divide in video comprehension. |
| Outcome: | The proposed video evaluation benchmark bridges cultural, linguistic, and domain divides . existing benchmarks only feature videos from YouTube, Shutterstock, or established video datasets based on cultural diversity . |
Variational Knowledge Graph Reasoning (N18-1)
Copied to clipboard
| Challenge: | Existing knowledge graphs have large amount of missing links, which limits their application . a recent study has proposed to design an automated inference model to complete the missing links in large knowledge graph. |
| Approach: | They propose to use variation inference to solve missing links in knowledge graphs . they use a posterior approximator, prior (path finder) and likelihood (path reasoner) |
| Outcome: | The proposed model achieves state-of-the-art on multiple datasets and is highly accurate. |
Open-Domain Question-Answering for COVID-19 and Other Emergent Domains (2021.emnlp-demo)
Copied to clipboard
| Challenge: | a system for open-domain question-answering is developed for COVID-19 . small data size allows system to retrieve answers from large corpus of scientific papers . |
| Approach: | They propose an open-domain question-answering system that can retrieve answers from large corpus of COVID-19 papers. |
| Outcome: | The proposed open-domain question-answering system can retrieve answers from large corpus of COVID-19 scientific papers. |