A method and system for predicting security defects in aviation software based on the fusion of LLM and dynamic knowledge.

Abstract

This invention discloses a method and system for predicting security defects in aviation software based on LLM and dynamic knowledge fusion. The method includes: collecting a general domain defect report dataset and an aviation domain defect report dataset; constructing a general domain security vocabulary based on the general domain defect report dataset; constructing an aviation domain security knowledge graph based on the aviation domain defect report dataset using a large language model and the Prompt project; fusing the general domain security vocabulary and the aviation domain knowledge graph to construct an aviation software security defect prediction model; dividing the aviation software defect reports to be predicted into training and testing sets to train the aviation software security defect prediction model; and using the trained aviation software security defect prediction model to complete the security prediction of the aviation software defect reports to be predicted. This invention effectively integrates knowledge from multiple domains, improving the accuracy of defect report analysis and its adaptability to the aviation domain.

Images