Reproduce “Human Cognition Inspired RAG with Knowledge Graph for Complex Problem Solving” 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

Yao Cheng, Yibo Zhao, Jiapeng Zhu, Yao Liu, Xing Sun, Xiang Li

CogGRAG, a human cognition-inspired framework from East China Normal University and Tencent Youtu Lab, enhances Large Language Models' complex problem-solving abilities by integrating structured knowledge graph data with a top-down decomposition and a dual-process self-verification mechanism. The system consistently achieved superior accuracy and reduced hallucination rates across various knowledge graph question answering benchmarks, demonstrating a more trustworthy approach.

Key insights

  • Modeling human cognitive strategies—specifically problem decomposition, hierarchical planning, and self-reflection—can effectively enhance LLM reasoning capabilities and reduce hallucinations in complex tasks.
  • Unifying problem decomposition, knowledge retrieval, and reasoning within a single, graph-structured cognitive framework allows for early integration of relational knowledge, preventing error propagation inherent in iterative RAG.
  • A dual-process self-verification mechanism, where a dedicated verifier LLM assesses the output of a reasoning LLM, significantly improves the factual consistency and reliability of generated answers.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.