Analysis And Implementation Of Spammers Detection Method Based On Social Network

With the flourish of the Web,social network has become a significant platform for people to share information and communicate with others.People expect social network to be safe and reliable.However,the endless dissemination of spam has caused great interference to normal users and may even threaten users' personal privacy.Thus spammer detection has become an urgent problem to be solved.In this thesis,two kind of algorithms have been proposed to detect spammers.One of them is based on content recognition and another is based on feature recognition.On the one hand,the convolution neural network CharCNN based on character-level is used for the spam detection.By using Chinese characters as input directly,the superiority of Chinese characters for character-level CNN is verified by experiments.The accuracy is 93%.On the basis of this,in order to locate spammer through the spam information,the sequential probability ratio model is introduced,and the message sent by the user is classified by the CNN model.The classification result sequence is used to the real-time monitoring and classification of the spam identification.On the other hand,taking Sina microblogging as the research object,this thesis prompts an available feature set for spammer classification using traditional classifiers by considering user information,user behaviors,social relations and blog text contents.Then this thesis analyzes available feature set and designs the feature information acquisition algorithm.Several feature selection algorithm is used for feature sorting.After the feature sorting,the combination of the optimal feature subset and the best classifier is selected by comparing performance of the different classifiers in the balanced data set and the unbalanced data set.And the experiment proves the effectiveness of the spammer detection algorithm,the accuracy is up to 90%.Finally,a spammer detection and labeling system is designed and implemented.The simulation results show the validity of the two algorithms.