Research On Event Participation Prediction Based On Distinguishing User Activity

Event-Based Social Network(EBSN)has developed rapidly in recent years,which provides users a platform to organize,participate and share offline events.It can not only help users find suitable events,but also help organizers find right participants by predicting whether a user will participate an event.The distribution of user participation fits a power-law distribution,the tail users are called as inactive users.These users participate in few events,but they are numerous,and may become active users.However,most of the existing event participation prediction methods can only predict the participation behaviors for the user who participated in a certain number of events before,while can't effectively predict that for inactive users.Few methods consider inactive users,but they can only achieve good results in case that social ties are close.To solve this problem,we propose a Double-layer Local Random Walk(DLRW)method,which looking for similar users in the user's local social circle by random walk,and extract features from these users,thus to impove the prediction performance of inactive users.In addition,the participation behaviors of active users and inactive users are different.However,the existing event participation prediction methods do not consider such difference.To solve this problem,we propose a Multi Factor event participation prediciton model for Different Activity Users(MFDAU),which extracting features from content,time,space and social relationships,then training different parameters for diffrent type of users to improve the overall prediction preformance effectively.In order to verify our model,we conduct experiments on Douban Event dataset and compare with some existing methods.The results show that our method performs well for different type of users.