A Knowledge Reasoning Modeling for security vulnerability databases based on Gaussian-SVM hybrid clustering

At present, network security becomes more and more important, but the security vulnerability data in various databases are relatively dispersed and it is difficult to establish the connection between them, network security knowledge graph can associate the security vulnerability data of different databases together, and it can be updated in real time with the latest security vulnerabilities. This paper establishes network security knowledge graph based on CAPEC, CWE and other security libraries. Aiming at the problem that TransR model cannot utilize entity semantic information, TransR-Roberta model is proposed, and splicing layer, attention layer and convolutional layer are added on the basis of TransR model. And for the problem that the quality of negative samples generated by Bernoulli negative sampling is too low, a clustering-based negative sampling optimization method is proposed. Firstly, the clustering task is accomplished with the help of Gaussian hybrid clustering algorithm on the entities in the security vulnerability knowledge graph to label the entities with clusters. In the next step, the clustered vulnerability entities are reclassified using support vector machine. Finally, a classifier combined with Facolloss function is added to the loss function of the TransR model, which is used to improve the ability of the knowledge inference model to distinguish between positive and negative samples. The experimental results show that the improved model improves the hit@N, MR and MMR metrics for the knowledge reasoning task.