Research And Application Of Web Malicious Generation Detection Technology Based On Deep Learning

Nowadays,with the vigorous development of the Internet,there are more and more various security threats appear on the Internet,and information leakage incidents occur more and more frequently.As part of the current network security issues,the detection of malicious code on web pages has become one of the research foucus of researcher.When the traditional machine learning method is used to detect JavaScript malicious code,there exists issues that the process of extracting features is intricate,with a great deal of calculations,and the code which is malicious confused and difficult to detect,indeed,which is not available in the requirements for the accuracy of JavaScirpt malicious code detection.Hence,a method based on deep learning,which is applied to detect JavaScript malicious code,is proposed in the paper,the feature extraction capability of JavaScript malicious code is improved.At the same time,the accuracy of JavaScript malicious code detection is improved and the false alarm rate is reduced.Due to JavaScript code has the characteristics of long-distance dependence,there are some defects in extracting the above dependence features while using the long-short-term memory network proposed by existing researchers.According to excellent achievements of RNN in the field of natural language processing,a bidirectional long-short term memory network model is proposed to extract the context dependent features of JavaScript malicious code to detect JavaScript malicious code.The experimental results show that,compared with the traditional detection methods,the method in this paper is more excellent in terms of detection rate and false alarm rate.The features extracted by the bidirectional long-short term memory network contain a lot of information that is not highly relevant to JavaScript malicious code.This issue leads to an increase in the amount of calculation and a low detection efficiency.Therefore,a detection model fused with attention mechanism is proposed,which is using the Encoder-Decoder architecture commonly used in deep learning.Encoder is composed of bidirectional long-short term memory network and Decoder is composed of long-short term memory network.Then the attention mechanism is used to calculate the attention weight of the input data on JavaScript malicious code,which extracting information related to the significance of JavaScript malicious code detection.At last,based on the above-mentioned BiLSTM detection model,a simple application is implemented to detect whether there is malicious code in a web page.