融合语义与结构特征的威胁情报文本攻击意图识别方法OACHSSCD

[Purpose]Threat intelligence texts provide important data support for security decision-making and intelligence analysis by tracking and analyzing key attack techniques and tactics.How to accurately identify attack intentions and tactical features from complex and unstructured text has become a key issue in promoting automated intelligence analysis and knowledge system construction.[Method]To this end,the study proposes a semantic structural fusion modeling method for threat intelligence texts.Firstly,to address the issues of se-vere unstructured threat intelligence text and sparse label distribution,a feature generation module centered on topic semantic modeling and context clustering was designed to enhance the separability of text expression.On this basis,the BERT pre-trained model is used to extract the contextual semantic vector representation of the text,and dimension level semantic interaction is achieved by introducing cross feature enhancement structure;Finally,to address the inherent issue of imbalanced class distribution in threat intelligence data,a class balanced Focal loss function is used to optimize the model training process.[Result/Conclusion]The experimental results show that the method exhibits excellent performance on real threat intelligence datasets,especially in multi label recognition tasks,with the highest F0.5 value reaching 92.63%,demonstrating good label modeling ability and practical application value.