Tracing Vulnerability to Attack Patterns Using Text Similarity

In order to quickly discover attack patterns that may be used against vulnerabilities and provide security personnel with more security knowledge and reference information, we propose a Weighted-SBERT (WSBERT) based text similarity calculation method to trace vulnerability to attack patterns. The WSBERT method take representative keywords into consideration to get more accurate sentence embeddings and calculate the similarities between attack patterns and vulnerabilities using sentence embeddings. To demonstrate the effectiveness of our proposed method, we compare the WSBERT method to the TF-IDF similarity method, and conduct ablation experiments. The results indicate that our WSBERT method performs better than the unimproved SBERT method, has similar performance to the TF-IDF method, and possesses stronger generalization ability and the ability to process long texts.