While users can interact with others online, more and more social networking services can help people to organize various offline social events, such as dinner parties and study groups, on the Internet. The hosts can invite friends or strangers to participate in their events in either manual or collaborative manner. However, such invitation manners may cost substantial time. Besides, the invitees may be uninterested or even unexpectedly contain spammers. In this paper, we aim at developing a predictive model to accurate recommend event participants. Specifically, given the host who initializes a social event, along with its event contexts, including the underlying social network, categories, and geolocations, our model will recommend a ranked list of candidate participants with the highest participation potential. We propose a feature-based matrix factorization model that optimizes pairwise errors of user rankings for training events, using six categories of features that represent the tendency of a user to attend the event. Experiments conducted on two event-based social networks Meetup and Plancast and Twitter retweet data exhibit the promising performance of our approach, together with an extensive study to analyze the factors affecting users’ event participation.
All Science Journal Classification (ASJC) codes
- Information Systems
- Human-Computer Interaction
- Hardware and Architecture
- Artificial Intelligence