知识增强文本语义特征的漏洞自动分类模型OA

In automatic vulnerability classification tasks,the uneven distribution of text samples across fine-grained vulnerability types and the significant differences in information contained in vulnera-bility descriptions pose major challenges.To address these issues,this paper proposes an automatic vul-nerability classification model with knowledge enhanced text semantic features,BerTrans H-knowledge-enhanced(BT H-KE).In the data preprocessing stage,CWE vulnerability types in the Alibaba Cloud vulnerability database are consolidated according to the Guidelines for Classification and Grading of Cybersecurity Vulnerabilities.In the knowledge-fused semantic feature enhancement stage,the model identifies distinctive features in vulnerability texts to represent related vulnerability attribute knowl-edge,and proposes a cross-dimensional fusion strategy that integrates vulnerability attribute knowledge with textual semantics.This strategy bridges the dimensional gap between knowledge and text represen-tation,enriching and strengthening key features of vulnerability texts.Experimental results show that the BTH-KE model achieves an accuracy of 93.37%,a precision of 93.55%,a recall of 93.37%,and an F1 score of 93.41%,respectively.