Reproduce “KARPA: A Training-free Method of Adapting Knowledge Graph as References for Large Language Model's Reasoning Path Aggregation” 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
Siyuan Fang, Kaijing Ma, Tianyu Zheng, Xinrun Du, Ningxuan Lu, Ge Zhang, Qingkun Tang
KARPA introduces a training-free method for Knowledge Graph Question Answering that leverages large language models to pre-plan complete reasoning paths, which are subsequently matched to the knowledge graph for answer aggregation. The method notably improves accuracy on datasets like WebQuestionSP and Complex WebQuestions, while reducing LLM-KG interactions by more than half compared to prior approaches.
Key insights
- Leveraging LLMs for global pre-planning of complete reasoning paths, rather than fragmented, step-by-step traversal, significantly enhances the coherence and accuracy of KGQA.
- A training-free semantic matching strategy across the entire KG, combined with LLM pre-planning, offers high adaptability across different LLMs and KGs without requiring specific model retraining.
- Heuristic value-based path matching effectively compares reasoning paths of varying lengths, enabling the identification of globally optimal paths and preventing entrapment in local optima.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.