Papers by Zhenhua Liu

23 papers
CART: A Generative Cross-Modal Retrieval Framework With Coarse-To-Fine Semantic Modeling (2025.acl-long)

Copied to clipboard

Challenge: Cross-modal retrieval tasks are used to retrieve data from one modality or another based on a query from another modality.
Approach: They propose a generative cross-modal retrieval framework based on coarse-to-fine semantic modeling . they propose combining K-Means and RQ-VAE to discretize multimodal data into token sequences that support autoregressive generation.
Outcome: The proposed framework achieves excellent performance and efficiency in multimodal retrieval tasks.
UAQFact: Evaluating Factual Knowledge Utilization of LLMs on Unanswerable Questions (2025.findings-acl)

Copied to clipboard

Challenge: Existing datasets to assess LLMs' performance on unanswerable questions lack factual knowledge support.
Approach: They propose a bilingual unanswerable question dataset with auxiliary factual knowledge created from a Knowledge Graph and two new tasks to measure LLMs' ability to utilize internal and external factual information.
Outcome: The proposed datasets show that LLMs do not consistently perform well even when they have factual knowledge stored.
How Pre-trained Language Models Capture Factual Knowledge? A Causal-Inspired Analysis (2022.findings-acl)

Copied to clipboard

Challenge: Recent studies show that pre-trained language models can fill in the missing factual words in cloze-style prompts such as ”Dante was born in [MASK]” .
Approach: They propose to quantitatively measure and evaluate the word-level patterns that PLMs depend on to generate the missing factual words.
Outcome: The proposed model fills in the missing factual words in cloze-style prompts by relying on effective clues or shortcut patterns.
MalURLBench: A Benchmark Evaluating Agents’ Vulnerabilities When Processing Web URLs (2026.findings-acl)

Copied to clipboard

Challenge: Existing models struggle to detect elaborately disguised malicious URLs, despite their ability to process malicious URL's.
Approach: They propose a benchmark to evaluate LLMs’ vulnerabilities to malicious URLs and a lightweight defense module to mitigate the vulnerability.
Outcome: The proposed framework analyzes 61,845 attack instances spanning 10 real-world scenarios and 7 categories of real malicious websites.
MINER: Multi-Interest Matching Network for News Recommendation (2022.findings-acl)

Copied to clipboard

Challenge: Existing methods learn a single user embedding from user’s historical behaviors to represent the reading interest.
Approach: They propose a poly attention scheme to learn multiple interest vectors for each user, which encodes the different aspects of user interest.
Outcome: The proposed approach significantly outperforms existing state-of-the-art methods on the MIND news recommendation benchmark.
Evolutionary Guided Decoding: Iterative Value Refinement for LLMs (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for directing language model outputs are limited in their accuracy due to a distributional gap . existing methods train static value functions on trajectories sampled exclusively from the base policy .
Approach: They propose a framework to bridge a distributional gap in the accuracy of value functions . they propose RLHF to align language models with human values and task requirements .
Outcome: The proposed framework reduces computational costs and improves value function accuracy by leveraging principled value function optimization.
Cocktail: A Comprehensive Information Retrieval Benchmark with LLM-Generated Documents Integration (2024.findings-acl)

Copied to clipboard

Challenge: Large Language Models (LLMs) have led to an influx of AI-generated content on the internet, transforming corpus of Information Retrieval (IR) systems from human-written to a coexistence with LLM-generated contents.
Approach: They propose a benchmark named Cocktail that compares IR models with LLMs to find relevant documents and passages from a corpus.
Outcome: The proposed benchmark aims to evaluate IR models in the mixed-sourced data landscape of the LLM era.
Jakiro: Boosting Speculative Decoding via Decoupled MoE (2026.acl-long)

Copied to clipboard

Challenge: Existing methods to accelerate large language model inference have a fundamental limitation: candidates at the same tree layer share identical feature representations, constraining diversity and diminishing overall effectiveness.
Approach: They propose a decoupled mixture of experts (MoE) into a draft model to generate diverse tokens from distinct feature spaces.
Outcome: The proposed approach achieves significant speedups over strong baselines, with notable improvements in non-greedy scenarios where token diversity is crucial.
Filtering before Iteratively Referring for Knowledge-Grounded Response Selection in Retrieval-Based Chatbots (2020.findings-emnlp)

Copied to clipboard

Challenge: Experimental results show that FIRE outperforms previous methods for building knowledge-grounded retrieval-based chatbots . a method called Filtering before iteratively referring is used to ground a conversation on background knowledge .
Approach: They propose a method for grounding conversation on background knowledge . they use context filter and knowledge filter to make context and knowledge aware . experimental results show that FIRE outperforms previous methods .
Outcome: The proposed method outperforms previous methods on two datasets.
Wider & Closer: Mixture of Short-channel Distillers for Zero-shot Cross-lingual Named Entity Recognition (2022.emnlp-main)

Copied to clipboard

Challenge: Existing mainstream methods for zero-shot cross-lingual named entity recognition ignore the rich and complementary information lying in the intermediate layers of pre-trained language models and domain-invariant information is easily lost during transfer.
Approach: They propose a mixture of short-channel distillers to fully interact the rich hierarchical information in the teacher model and to transfer knowledge to the student model sufficiently and efficiently.
Outcome: The proposed method shows great generalization and compatibility across languages and fields.
DiffusionDialog: A Diffusion Model for Diverse Dialog Generation with Latent Space (2024.lrec-main)

Copied to clipboard

Challenge: Existing studies have tried to introduce discrete or Gaussian-based latent variables to address the one-to-many problem, but the diversity is limited.
Approach: They propose a diffusion model to enhance the diversity of dialogue generation by using continuous latent variables instead of discrete ones.
Outcome: The proposed model greatly enhances diversity of dialog response while keeping the coherence.
LICHEE: Improving Language Model Pre-training with Multi-grained Tokenization (2021.findings-acl)

Copied to clipboard

Challenge: Pre-trained language models are trained based on single-grained tokenization, making it hard to learn the precise meaning of coarse-grain words and phrases.
Approach: They propose a language model pretraining method that incorporates multi-grained information of input text into pre-trained language models.
Outcome: The proposed method improves performance on CLUE and SuperGLUE in Chinese and English with little extra inference cost.
Learning to Refuse: Towards Mitigating Privacy Risks in LLMs (2025.coling-main)

Copied to clipboard

Challenge: Large language models (LLMs) exhibit remarkable capabilities in understanding and generating natural languages, but can inadvertently memorize private information, posing significant privacy risks.
Approach: They propose to use a dataset to evaluate machine unlearning methods for protecting personal data in a realistic scenario.
Outcome: The proposed model outperforms baseline methods by 5.65 points and protects target individuals’ personal data while maintaining general capabilities.
Text Classification by Contrastive Learning and Cross-lingual Data Augmentation for Alzheimer’s Disease Detection (2020.coling-main)

Copied to clipboard

Challenge: Existing methods for AD detection are too expensive and time-consuming to cover all potential patients.
Approach: They propose a contrastive learning method to obtain effective text representations based on monolingual embeddings of BERT and a cross-lingual data augmentation method by building autoencoders to learn the text representation shared by both languages.
Outcome: The proposed method outperforms other methods on a Mandarin AD corpus and achieves 81.6% detection accuracy.
Draft, Verify, Restore: Self-Refining Historical Inscription Restoration with a Unified MLLM (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for end-to-end historical inscription restoration rely on task-separated pipelines with irreversible error accumulation and patch-based generation that sacrifices page-level consistency.
Approach: They propose a unified MLLM for end-to-end historical inscription restoration that integrates draft-guided localization and Hierarchical self-refinement to enable accurate damage localization.
Outcome: The proposed model achieves superior performance in both text restoration accuracy and appearance restoration quality.
MM-Doc-R1: Training Agents for Long Document Visual Question Answering through Multi-turn Reinforcement Learning (2026.findings-acl)

Copied to clipboard

Challenge: Existing work on long document visual question answering is based on Retrieval-Augmented Generation (RAG) where textual or visual content is encoded into embeddings and relevance is determined by similarity scores with respect to the original query.
Approach: They propose a framework that employs an agentic, vision-aware workflow to address long document visual question answering through iterative information discovery and synthesis.
Outcome: The proposed framework outperforms existing RL systems by 10.4% on the MMLongbench-Doc benchmark and demonstrates superior training performance over GRPO.
ICG: Improving Cover Image Generation via MLLM-based Prompting and Personalized Preference Alignment (2025.emnlp-main)

Copied to clipboard

Challenge: Large language models and diffusion models have opened new possibilities for AI-generated content . personalized cover image generation remains underexplored despite its critical role in boosting user engagement on digital platforms.
Approach: They propose a framework that integrates MLLM-based prompting with personalized preference alignment to generate high-quality, contextually relevant covers.
Outcome: The proposed framework improves image quality, semantic fidelity, and personalization, leading to stronger user appeal and offline recommendation accuracy in downstream tasks.
Reviving Cultural Heritage: A Novel Approach for Comprehensive Historical Document Restoration (2025.acl-long)

Copied to clipboard

Challenge: Existing methods for historical document restoration focus on single modality or limited-size restoration, failing to meet practical needs.
Approach: They propose a full-page HDR dataset and an automated HDR solution to replace manual restoration methods.
Outcome: The proposed solution improves OCR accuracy from 46.83% to 84.05% when processing severely damaged documents, with enhancement to 94.25% through human-machine collaboration.
DiffStyleTTS: Diffusion-based Hierarchical Prosody Modeling for Text-to-Speech with Diverse and Controllable Styles (2025.coling-main)

Copied to clipboard

Challenge: Existing models for text-to-speech (TTS) synthesize speech with acoustic features . autoregressive models have problems with word skipping and repeated reading . non-autoregressive acustic models lack probabilistic modeling and unimodal characteristics of Gaussian distribution don't conform to true distribution of aural features, which restricts the diversity of generated prosodic features.
Approach: They propose a multi-speaker acoustic model that hierarchically models speech prosodic features and controls different prosodic styles to guide prosody prediction.
Outcome: The proposed method outperforms baseline models in naturalness and achieves superior synthesis speed compared to baseline models.
RecBase: Generative Foundation Model Pretraining for Zero-Shot Recommendation (2025.emnlp-main)

Copied to clipboard

Challenge: Existing methods for addressing item-level user interests are lacking in cross-domain generalization . RecBase model is domain-agnostic and can be used to enhance recommender systems' effectiveness .
Approach: They propose a domain-agnostic foundational model pretrained with a recommendation-oriented objective that leverages a large-scale, heterogeneous, cross-domain corpus with unified textual representations and feature mappings to enhance cross- domain generalization.
Outcome: The proposed model matches or surpasses baselines in zero-shot and cross-domain recommendation tasks on eight real-world datasets.
GIFT: Graph-Induced Fine-Tuning for Multi-Party Conversation Understanding (2023.acl-long)

Copied to clipboard

Challenge: Existing methods on understanding multi-party conversations typically embed interlocutors and utterances into sequential information flows or use superficial graph structures.
Approach: They propose a plug-and-play method which adapts Transformer-based pre-trained language models for universal MPC understanding.
Outcome: The proposed method can adapt Transformer-based pre-trained language models for universal MPC understanding.
Probing Language Models for Pre-training Data Detection (2024.acl-long)

Copied to clipboard

Challenge: Large Language Models (LLMs) have shown impressive capabilities, while raising concerns about the data contamination due to privacy issues and leakage of benchmark datasets in the pre-training phase.
Approach: They propose to utilize the probing technique to examine the model’s internal activations to detect pre-training data contamination by examining the model's internal activates.
Outcome: The proposed method outperforms baselines and achieves state-of-the-art performance on both WikiMIA and ArxivMIA, with additional experiments confirming its efficacy.
Detecting Speaker Personas from Conversational Texts (2021.emnlp-main)

Copied to clipboard

Challenge: Existing studies on personas are pre-defined and hard to obtain before a conversation . a new task aims to detect speaker persona based on conversational text .
Approach: They propose a task to detect speaker personas based on conversational text . they build a dataset for SPD and propose utterance-to-profile matching networks .
Outcome: The proposed task outperforms baseline models and utterance-to-profile (U2P) matching networks.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations