I²Rec: An Iterative and Interactive Recommendation System for Event-Based Social Networks

Abstract

Event-based social networks (EBSNs) such as Meetup and Plancast have been emerging in recent years. In addition to the online virtual groups and connections on many existing Online Social Networks (OSNs), EBSNs also provide a platform for users to initialize and manage offline physical events in which user’s activities are strongly geographically constrained. As an increasing number of users are attracted by EBSNs, it is highly desirable to provide users with accurate recommendations of both online groups and offline events, which has become an urgent need. In this paper, we propose a comprehensive study for recommending users online groups and offline events on EBSNs. To represent user’s interactions via a timeline horizon, we design an I terative and I nteractive Rec ommendation System (I $^2$ Rec), which couples both online user activities and offline social events together for providing more accurate recommendation. Our proposed $I^2Rec$ system infers a user’s online and offline activities in turn and iteratively enriches the training information based on user’s feedback. Using the large-scale real-world dataset crawled from Meetup, our recommendation system outperforms other baseline approaches significantly. More importantly, the empirical results also validate that our proposed system can continuously provide accurate recommendation over time by capturing users’ changing interests.