Reproduce “GraphRAG-LTM: A Layered Threat Modeling Framework for Attacks on Graph Retrieval-Augmented Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yining Tang, Junyu Zhang, Haomiao Qiu, Yongchao Liu, Jiajun Zheng, Zhao Wang, Ziyue Qiao, Chao Wu
GraphRAG-LTM examines attacks targeting GraphRAG through query changes and edits to query-specific evidence graphs. It compares black-box query attacks using ranking feedback with white-box attacks using retriever gradients, then adds bridge-prioritized triplet deletion. On HotpotQA-Distractor with Contriever and LLaMA3-8B, Answer F1 falls from 52.3 clean to 37.6, 26.2, and 23.7; these results apply to evaluated settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.