Abstract: Aiming at the problem that malicious Web page links and plugins could be attached to other files through constant confusion and deformation, traditional detection methods were difficult to achieve accurate detection, we proposed a Web page attack redirection confusion detection method based on multimodal deep neural networks. Firstly, we extracted the features of  Web page attacks: attribute class, keyword class, var class, and word class, and converted them into 8-dimensional sensitive feature vectors to 
calculate their corresponding real values. Secondly, the Web page and real values were input together into a multimodal deep neural network for training. Finally, accurate attack redirection confusion detection results were obtained through the output of the Web page classifier. The experimental results show that the detection rate of the proposed method is about 98%, which can effectively detect redirection confusion in Web page attacks while ensuring a high detection rate.

Key words: multimodal deep neural network, Web page attack redirection confusion detection, TF-IDF algorithm, nonlinear excitation unit, loss function