JavaScript Obfuscation Detection Methods Based On CNN

At present,the Internet is bringing great convenience to us,but at the same time,the malicious web attacks occur frequently,most of these malicious web pages are implemented through the JavaScript obfuscation.In order to fight against malicious JavaScript codes and protect the safety of the users' personal property,detection of JavaScript obfuscation codes has become an important field of the current.The traditional JavaScript obfuscation codes detection method mainly adopts shallow machine learning models or metric methods,and all of these require prior in-depth study of JavaScript obfuscation codes.Besides,they are strict in feature extraction,which directly determine the precision of detection.Therefore,a character matrix feature extraction method based on Bigram was presented to simplify the process of feature extraction and a method of JavaScript obfuscation codes detection based on convolution neural network was presented to improve the precision.First of all,traditional JavaScript obfuscation codes feature extraction methods have several deficiencies such as complexity in process,the need for a long time to study JavaScript,lack of universality and so on.Therefore,a character matrix feature extraction method based on Bigram inspired by the Bigram semantic model and the Markov probability matrix was proposed to overcome these shortcomings.Secondly,most of the traditional machine learning detection methods are shallow models.For the characteristics of high requirements and low detection precision,a JavaScript obfuscation codes detection method based on convolution neural network was proposed.It utilizes the powerful learning ability of the convolution neural network in the classification process and not only reduces the requirement for the features but also effectively improves the precision of detecting JavaScript obfuscation codes.Finally,the experiments of the two methods given before are made.The results show that the character matrix feature extraction method proposed in this thesis has a certain ability to distinguish between normal and obfuscation JavaScript codes and the time that took is half of the traditional method.At the same time,compared with the traditional machine learning detection method,there is a better precision rate in method based on CNN and it improves 0.6%percent?...