Event Recommendation via Collective Matrix Factorization with Event-User Neighborhood

Abstract

Event-based social networks (EBSNs) recently emerge as a new type of social network and have been growing rapidly. Because of the very large volume of various events, the demand of event recommendation becomes increasingly important. In this paper, we propose a novel approach called Collective Matrix Factorization with Event-User Neighborhood (CMF-EUN) model to handle this problem. CMF-EUN combines the strengths of matrix factorization and neighborhood based methods. Due to the fact that RSVP matrix is generally extremely sparse, it is difficult to find similar neighborhoods using the widely adopted similarity measures. To address this, we calculate the similarities based on some specific features of events and users in EBSNs. The heterogeneous social relationships are also taken into consideration. Experimental results conducted on real datasets collected from DoubanEvent show that the proposed model provides superior performance and outperforms several baseline methods.