Improved Based On A Semi-supervised Sparse AutoEncoder IM Traffic Identification Model Of Comparison And Research

The accurate identification of traffic can not only do a big improvement of Internet working condition,but also accurate analysis and control over behavior according to the needs,complete the construction of user portraits and the personalized recommendation of website pages,which has strong market application value.However,There are not only many types of applications on the market today,but some applications take the form of encrypted transmissions such as SSL,based on present techniques,it's hard to identify application flow.At the same time,the current study of traffic identification for instant-messaging classes is relatively small in the industry.This paper first introduces the current main methods of traffic identification and researches both at home and abroad,and analyzes its existing problems.Then,this paper realizes the process of semi-automatic capturing of data packets to the full automatic capture of messages,The use of data processing module to complete the data cleaning work,and finally through the construction of an effective classification model to complete the APP identification,In view of this situation,this paper use the neural network algorithm of the SAE feature extraction model,with feature training for payload data and adding identification tags,effectively solve the present several popular IM flow classification problem.The recognition rate is high in recognition of accuracy.Meanwhile,compare with the Apriori feature extraction algorithm in complexity of space and time,we can find the optimal IM traffic identification model.So as to complete the construction of user portraits and the personalized recommendation of website pages,which has strong market application value.