GraphSearch: An Agentic Deep Searching Workflow for Graph Retrieval-Augmented Generation

Cehao Yang, Xiaojun Wu, Xueyuan Lin, Chengjin Xu, Xuhui Jiang, Yuanliang Sun Jia Li, Hui Xiong, Jian Guo · 2025

Read GraphSearch: An Agentic Deep Searching Workflow for Graph Retrieval-Augmented Generation by Cehao Yang, Xiaojun Wu, Xueyuan Lin, Chengjin Xu, Xuhui Jiang, Yuanliang Sun Jia Li, Hui Xiong, Jian Guo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-22009.txt.

Description

Cehao Yang and eight co-authors present GraphSearch in this 2025 paper, a deeper way for language models to search knowledge graphs before answering. Graph-based retrieval grounds answers in structured facts, but current systems search shallowly and miss key evidence. They also use their pre-built graph data poorly. The authors aim for a search that reasons as it goes.

GraphSearch organizes retrieval into six modules that interact over several rounds, refining the search as partial answers form. It also runs two channels at once: meaning-based queries over text chunks, and relation-based queries over the graph structure. The two channels cover for each other's blind spots. Across six benchmarks that need multi-step reasoning over sources, the workflow improved both answer accuracy and generation quality over the standard strategy. The authors offer it as a promising direction for graph-based retrieval.

Catalog record

CREATORCehao Yang, Xiaojun Wu, Xueyuan Lin, Chengjin Xu, Xuhui Jiang, Yuanliang Sun Jia Li, Hui Xiong, Jian Guo
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-22009

Browse the full library · About the Tower of Babel