Research On Technology Of Malicious Users Identification Based On Weibo Content

Since the arrival of the media era,more and more people have joined the life of network.People share their own activities and browse a variety of new things on social platform.But the social platform has also attracted a lot of spammers who publish various advertisements or other spams that seriously affect the experience of users while using the platform.Sina Weibo as one of the most popular Chinese social platform is also affected by a great amount of spammers,thus to detect different spammers is very necessary.This thesis proposes a method that uses semantic analysis,statistical analysis and machine learning to detect different spammers based on contents which were published by users.This method can effectively do multi-class detection of spammers in Sina Weibo.The results are as follows.Firstly,implementing a scalable malicious dictionary that can recognize malicious words variants.This thesis improves the dictionary based on the semantic information by adding stop-words table extracted from Weibo environment and considering malicious words variants.The dictionary can be used to complete malicious comments detection experiment.The results show that under the premise of guaranteeing the accuracy of malicious comment detection,the recall rate can be improved and reaches to 82.8%.Secondly,designing and implementing a method of using statistical information to distinguish users.This thesis uses the malicious dictionary to calculate the malicious scores of users in Sina Weibo.According to these scores and the method,users are divided into three categories,namely,ordinary users,advertising users and voting users.Each user's recall rate and accuracy can reach 90%.Finally,achieving the malicious user detection based on machine learning.This thesis uses the malicious dictionary to extract some eigenvalues,such as,malicious contents proportion and posting interval.Then decision tree algorithm,AdaBoost algorithm and SVM algorithm are used to distinguish users.The results show that the performance of SVM algorithm is best.