MFDAU: A multi-features event participation prediction model for users of different activity levels

Event-based social networks (EBSNs) provide a platform for users to organize, participate in and share events. Predicting event participants in EBSNs is beneficial for many applications, such as event recommendation and influence analysis. Existing event participation prediction methods do not consider the differences between active users and inactive users, resulting in the overall performance for users of different activity levels is not good enough. For this problem, we propose a multi-features model for users of different activity levels (MFDAU) to improve the overall performance of event participation prediction by extracting user features from multiple aspects, including event content, time, space and social relation, and training different feature parameters for active users and inactive users respectively. Besides, considering it is more challenging to predict event participation for inactive users, we propose a double-layer local random walk (DLRW) method to extract the social features of users, which is expected to improve the prediction performance for inactive users by finding similar users from one user’s local social circle and extracting effective user features according to these similar users. Experimental results on real dataset collected from DoubanEvent show that our proposed model is superior to several comparison schemes, such as MFN, PEA and DSI, which increases the average F1-score for overall users by 3.9%, 17.3% and 6.5%, active users by 3.5%, 17.3% and 10.9%, and inactive users by 4.6%, 17.7% and 6.2% respectively.