bubblerag-evidence-driven-graphs

【效率与协作】Address hallucinations in LLM QA over black-box knowledge graphs using evidence-driven retrieval. Formalize Optimal Informative Subgraph Retrieval and employ bubble expansion to discover candidate evidence graphs, achieving state-of-the-art multi-hop QA performance.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准