XSS Attack Detection Based on Multisource Semantic Feature Fusion DOI Open Access

Ze Hu,

Jianwei Zhang, Hongyu Yang

и другие.

Electronics, Год журнала: 2025, Номер 14(6), С. 1174 - 1174

Опубликована: Март 17, 2025

Cross-site scripting (XSS) attacks can be implemented through various attack vectors, and the diversity of these vectors significantly increases overhead required for detection systems. The existing XSS methods face issues such as insufficient feature extraction capabilities attacks, inadequate multisource fusion processes, high resource consumption levels their models. To address problems, we propose a novel approach based on semantic fusion. First, design normalized tokenization rule structural features code use word embedding model to generate original XSS. Second, local network depthwise separable convolution (DSC) that extracts text syntactic using kernels with different sizes. Then, bidirectional long short-term memory (Bi-LSTM) extract global Finally, introduce multihead attention employs saliency score dynamic weight adjustment mechanism identify key parts input sequence dynamically adjust each head. This enables deep features. Experimental results demonstrate proposed achieves an F1 99.92%, outperforming methods.

Язык: Английский

XSS Attack Detection Based on Multisource Semantic Feature Fusion DOI Open Access

Ze Hu,

Jianwei Zhang, Hongyu Yang

и другие.

Electronics, Год журнала: 2025, Номер 14(6), С. 1174 - 1174

Опубликована: Март 17, 2025

Cross-site scripting (XSS) attacks can be implemented through various attack vectors, and the diversity of these vectors significantly increases overhead required for detection systems. The existing XSS methods face issues such as insufficient feature extraction capabilities attacks, inadequate multisource fusion processes, high resource consumption levels their models. To address problems, we propose a novel approach based on semantic fusion. First, design normalized tokenization rule structural features code use word embedding model to generate original XSS. Second, local network depthwise separable convolution (DSC) that extracts text syntactic using kernels with different sizes. Then, bidirectional long short-term memory (Bi-LSTM) extract global Finally, introduce multihead attention employs saliency score dynamic weight adjustment mechanism identify key parts input sequence dynamically adjust each head. This enables deep features. Experimental results demonstrate proposed achieves an F1 99.92%, outperforming methods.

Язык: Английский

Процитировано

0