Reproduce “Graph Counselor: Adaptive Graph Exploration via Multi-Agent Synergy to Enhance LLM Reasoning” 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
Junqi Gao, Xiang Zou, YIng Ai, Dong Li, Yichen Niu, Biqing Qi, Jianxing Liu
Graph Counselor is a Graph Retrieval-Augmented Generation (GraphRAG) framework that addresses limitations in current methods regarding information aggregation and reasoning adaptivity. The framework, employing a multi-agent collaborative extraction module and a self-reflection mechanism, demonstrated up to a 24.2% improvement in Rouge-L scores over state-of-the-art Graph-CoT.
Key insights
- The Adaptive Graph Information Extraction Module (AGIEM)'s Planning Agent and the flexible operations of the Execution Agent are critical for effective graph information processing.
- The Self-Reflection with Multiple Perspectives (SR) module consistently improves performance, with two reflection iterations providing optimal results.
- Metacognitive abilities for self-reflection do not necessarily scale linearly with model size, allowing smaller models to be effective for the reflection process.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.