Qichuan LIU

Biography

I am currently a first-year Ph.D. student at the School of Informatics, Xiamen University, fortunate to be co-advised by Prof. Zhihong Zhang and Prof. Qinggang Zhang (Jilin University). In 2023, I received a Bachelor of Management degree from South-Central Minzu University. In Oct. 2024, I founded the GenIRAG Team and have close work with Chentao Zhang, Chenfeng Zheng, Yuxuan Hu, Tianke Xiang, and Zerui Chen.

My research interests include Information Retrieval (IR), Retrieval-Augmented Generation (RAG), Knowledge Graph (KG), and LLM-based Agent. My research topic is Trustworthy Retrieval-Augmented AI and Its Application, with recent work published in top-tier AI conferences including ICML'26, WWW'26, and ACL'25.

Publications(† for co-first author, ‡ for corresponding author)

ICML'26 From Retrieval to Translation: Translating Query into Graph-level Clues for Retrieval-Augmented Generation

Qichuan Liu†, Qinggang Zhang†, Yuxuan Hu, Chenfeng Zheng, Zerui Chen, Chentao Zhang, and Zhihong Zhang‡ · 2026

CCF A Graph Retrieval-Augmented Generation Graph Construction Constrained Decoding

Existing structure-augmented RAG systems are currently facing two key challenges: potential retrieval suspension and cumulative semantic drift. We propose KG-Translator, which constructs a novel graph (ParseKG) by leveraging a lightweight model for entity extraction and syntactic parsing. On top of ParseKG, a specialized LLM leverages the KG-constrained index to faithfully translate queries into graph-level clues for precise retrieval.

Details Paper (Coming Soon) Code & Data (Github)

WWW'26 Facilitating Generative Retrieval with Logical Denoising for Interpretable Conversational Search

Qichuan Liu, Chentao Zhang, Yuxuan Hu, Chenfeng Zheng, Qinggang Zhang, and Zhihong Zhang‡ · 2026

CCF A Conversational Search Generative Retrieval Long Chain-of-Thought

Conversational search faces noisy contexts and poor interpretability. We propose LogiCGR to equip LLMs with logical denoising and generative retrieval via curriculum learning and GRPO, with seamless integration in an adaptive framework. In addition, we introduce a self-dual-path retrieval module to yield complementary gains.

Details Paper (PDF) Code & Data (GitHub)

ACL'25 Beyond the Answer: Advancing Multi-Hop QA with Fine-Grained Graph Reasoning and Evaluation

Qichuan Liu, Chentao Zhang, Chenfeng Zheng, Guosheng Hu, Xiaodong Li, and Zhihong Zhang‡ · 2025

CCF A Multi-hop QA Question Decomposition Evaluation

The flawed reasoning processes in multi-hop QA can yield correct answers. We introduce PER architecture, which offers explicit intermediate reasoning steps for multi-hop QA. Moreover, we propose a PSE metric to evaluate intermediate reasoning steps from both planning and solving perspectives. With our PER-PSE framework, we find that RAG systems exhibit "fortuitous reasoning continuance" and "latent reasoning suspension".

Details Paper (PDF) Code (GitHub) Benchmark (HuggingFace)

Educations

• 2025.09-Now · Ph.D. Student, School of Informatics, Xiamen University, Xiamen, China.
• 2023.09-2025.06 · Master Student, School of Informatics, Xiamen University, Xiamen, China. (Proceed to Ph.D.)
• 2019.09-2023.06 · B.Mgt., School of Management, South-Central Minzu University, Wuhan, China.

Experiences

• 2023.03-2023.04 · China Electric Power Research Institute, State Grid, Beijing, China.

Honors & Awards

• 2023.07 · Official Media Report of South-Central Minzu University
• 2020-2023 · National scholarship for Undergraduates ~ 3 times
• 2022.07 · National Second Prize 🥈 in the 15th Chinese Collegiate Computing Competition
• 2022.06 · National First Prize 🥇 in the 10th "TipDM Cup" Data Mining Challenge
• 2022.03 · National Third Prize 🥉 in the 17th "Challenge Cup" Competition

Academic Services

• 2024 · Served as an invited reviewer at CIKM'24.