Publications

A collection of my research work. † denotes equal contribution.

LISRec: Modeling User Preferences with Learned Item Shortcuts for Sequential Recommendation

LISRec: Modeling User Preferences with Learned Item Shortcuts for Sequential Recommendation

Haidong Xin, Zhenghao Liu, Sen Mei, Ohters

KDD 2026

This work proposes LISRec, a framework that extracts personalized semantic shortcuts from user histories to filter noise and capture stable preferences, significantly improving sequential recommendation performance.

Code
ClueAnchor: Clue-Anchored Knowledge Reasoning Exploration and Optimization for Retrieval-Augmented Generation

ClueAnchor: Clue-Anchored Knowledge Reasoning Exploration and Optimization for Retrieval-Augmented Generation

Hao Chen, Yukun Yan, Sen Mei, Wanxiang Che, Others

EMNLP 2025

This work introduces ClueAnchor, a clue-anchored reasoning framework that generates and optimizes multiple reasoning paths to help RAG systems better extract and integrate key evidence, yielding more complete and robust reasoning even under noisy or implicit retrieval.

Code
UltraRAG v2: A Low-Code MCP Framework for Building Complex and Innovative RAG Pipelines

UltraRAG v2: A Low-Code MCP Framework for Building Complex and Innovative RAG Pipelines

Sen Mei, Haidong Xin, Chunyi Peng, Yukun Yan, Others

OpenBMB 2025

A Low-Code MCP Framework for Building Complex and Innovative Retrieval-Augmented Generation systems.

Code
RAG-DDR: Optimizing Retrieval-Augmented Generation Using Differentiable Data Rewards

RAG-DDR: Optimizing Retrieval-Augmented Generation Using Differentiable Data Rewards

Xinze Li†, Sen Mei†, Zhenghao Liu, Yukun Yan, Others

ICLR 2025

RAG-DDR is an end-to-end method that aligns RAG modules through data rewards, enabling LLMs to use retrieved knowledge more effectively than SFT.

Code
MARVEL: unlocking the multi-modal capability of dense retrieval via visual module plugin

MARVEL: unlocking the multi-modal capability of dense retrieval via visual module plugin

Tianshuo Zhou†, Sen Mei†, Xinze Li, Zhenghao Liu, Others

ACL 2024

MARVEL learns an embedding space for queries and multi-modal documents to conduct retrieval and encodes queries and multi-modal documents with a unified encoder model, which helps to alleviate the modality gap between images and texts. Specifically, we enable the image understanding ability of the well-trained dense retriever, T5-ANCE, by incorporating the visual module's encoded image features as its inputs. To facilitate the multi-modal retrieval tasks, we build the ClueWeb22-MM dataset based on the ClueWeb22 dataset, which regards anchor texts as queries, and extracts the related text and image documents from anchor-linked web pages. Our experiments show that MARVEL significantly outperforms the state-of-the-art methods on the multi-modal retrieval dataset WebQA and ClueWeb22-MM. MARVEL provides an opportunity to broaden the advantages of text retrieval to the multi-modal scenario. Besides, we also illustrate that the language model has the ability to extract image semantics and partly map the image features to the input word embedding space.

Code
Text matching improves sequential recommendation by reducing popularity biases

Text matching improves sequential recommendation by reducing popularity biases

Zhenghao Liu†, Sen Mei†, Chenyan Xiong, Others

CIKM 2023

This work introduces TASTE, a text-matching sequential recommendation model that verbalizes items and interactions while using sparse attention to model long histories, achieving state-of-the-art performance and alleviating cold-start and popularity-bias issues.

Code