Term-extract-enhanced Python-programming Question Answering with GraphRAG

Integrating local domain knowledge bases into domain-specific Question Answering (QA) systems enhances their professionalism and effectiveness. Recently, the Graph-based Retrieval-Augmented Generation (GraphRAG) framework has become a preferred approach for integrating local knowledge bases with large language models (LLMs). However, the standard GraphRAG exhibits limitations in extracting technical terms in domain-specific QA tasks, and often introduces irrelevant knowledge, leading to unfaithful answers. To address these issues, we develop a GraphRAG-based QA platform with redesigning prompt templates, which is specifically for College Cybersecurity Python Programming course. We first process various course documents to construct the local knowledge base. Then we deploy GraphRAG for the QA platform and design the prompt templates for technical term extraction and faithful answer generation. Experimental results demonstrate that the proposed QA platform significantly outperforms traditional methods in terms of accuracy and faithfulness.