Research On Abnormal User Detection In Social Networks

With the development of the Internet,social networks have affected all aspects of people's life,learning,work and entertainment.The diversification trend of the network makes the way of information dissemination more and more abundant.Social networking not only extends offline social activities to online,making it easier for users to communicate with each other,but also enables users to display themselves and get hot information and topics of interest through social platforms.However,in order to gain profits,some malicious users just use the advantages of social network platform to disseminate some malicious information,such as pornographic content,spam advertising,phishing and so on.These malicious users spread a large number of malicious messages through false accounts or by embezzling normal users' accounts.This malicious behavior seriously affects users' online experience and property security.Nowadays,the overflow of abnormal users has become a serious problem facing the social media service industry.Therefore,anomaly detection in social networks has always been a hot research area for scholars and business circles in recent years.This paper takes Sina Weibo social network platform as the research object,launches the research on abnormal user detection of Sina Weibo platform,mainly completes the following three aspects,and its theoretical methods are also applicable to other social network platforms.In order to obtain the ideal experimental data source,we use Scrapy crawler framework to design the personalized web crawler system for normal and abnormal micro-blog users,which can collect and import the data of users' micro-blog content,user information and user relationship efficiently into MongoDB database in real time,and make use of abnormal user reviews.And we use anomaly user evaluation criteria to construct anomaly user detection experimental data set.(2)By exploring the characteristics of user information and behavior of normal and abnormal users on Sina Weibo,this paper deeply analyses the data of normal and abnormal users,extracts new features with domain knowledge,and preprocesses data such as feature extraction and data formatting according to needs.The feature selection algorithm is used to rank all features in order to find out the important factors affecting anomalous user detection results.An anomalous user detection model based on user characteristics is constructed.The random forest algorithm with weighted voting is used as a training and detection model,and the model is compared with other algorithms.(3)The selection of parameters in random forest model is very important to the performance of the model.Traditional methods usually select parameters based on empirical values,which can not determine whether the ideal effect of the experiment can be achieved.In order to determine the optimal parameters of the model,we introduce standard particle swarm optimization(SPSO)to optimize the parameters of the model,and construct a weighted voting random forest anomaly user detection model based on SPSO optimization,which improves the detection performance of the model significantly.