LLM Reasoning Enhancement with Subgraph Generation

This paper addresses the limitations of Large Language Models (LLMs) in complex reasoning tasks by introducing a framework called SGR (Stepwise reasoning enhancement framework based on external subgraph generation). The core idea is to leverage external knowledge bases to create relevant subgraphs, guiding the LLM's reasoning process step-by-step over this structured information. This approach aims to mitigate the impact of noisy information and improve reasoning accuracy, which is a significant challenge for LLMs in real-world applications.