Analysis And Prediction Of User Departure Behavior Based On K-core In Dynamic Social Network

With the rapid advancement of social networks,various social networking sites and application software have emerged and become widely used,such as Facebook,You Tube,and Wechat(Whats App),which enables people to easily and quickly share information and communicate with users around the world.Currently,as the most mainstream communication medium,these social platforms attract a large number of users,which means that a great deal of interactive objects also floods into social networks.However,social networks are often dynamic in reality,with objects and links change dramatically over time.In recent years,people have done some in-depth research on dynamic social networks,such as the community detection,the most influential,and the link prediction.Generally,there are a large number of communities in a social network that can be regarded as a collection of users with similar interests,topics,and needs.However,as time passes,those communities have to face the loss of users and user departure.It will not make a significant impact on the use and healthy development of the platform if the user leaves temporarily or only a small number of users leave.On contrast,once a majority of users choose to leave from the platform,it will cause severe outcomes for the use and development of the platform.To detect this situation in advance and maintain the normality of the social network,that is extremely important to predict the departure of the node.In order to accurately predict the situation of the user's departure and maintain the regular operation of the social network,this paper mainly proposes an algorithm of user departure behavior prediction for the group.The algorithm takes into account the positive effects of group activity and the adverse effects of inactive friends in the group.It combines this information to predict whether the user is leaving.This paper also proposes an algorithm of user departure behavior prediction on social networks.This algorithm introduces the concept of benefits and puts forward global benefits and local benefits to calculate the impact of the entire network and neighbors.To achieve such an accurate prediction,this paper designs a time-based sliding window model,which enables data acquisition and accuracy comparison by window sliding.Finally,it is necessary to state that the experiments presented in this paper have beenvalidated on several real data sets.