Implementation of Multi-Level RAG Model for Enhanced Synergistic Vulnerability Analysis

Traditional vulnerability analysis models often rely on a single data source, which limits their ability to comprehensively detect and analyze vulnerabilities. In this study, we propose a multi-level Retrieval-Augmented Generation (RAG) model that progressively integrates multiple vulnerability databases to overcome these limitations. By incorporating core and peripheral data sources at different levels, our approach aims to enhance the quality of vulnerability analysis. We integrate four major vulnerability databases-National Vulnerability Database, CERT Vulnerability Notes Database, GitHub Advisory Database, and Exploit Database-into a Level 4 RAG model. This integration demonstrates an average performance improvement of 6.4% compared to a Level 1 model that only refers to the National Vulnerability Database. The evaluation is based on several metrics, including METEOR, BERT Score, Faithfulness, and Answer Relevancy, which collectively highlight the benefits of a multi-source, incremental integration strategy for more effective vulnerability analysis. These results highlight the potential of multi-level RAG models in enhancing vulnerability analysis through the integration of complementary vulnerability data sources.