Exploring social approach to recommend talks at research conferences

This paper investigates various recommendation algorithms to recommend relevant talks to attendees of research conferences. We explored three sources of information to generate recommendations: users' preference about items (i.e. talks), users' social network and content of items. In order to find out what is the best recommendation approach, we explored a diverse set of algorithms from non-personalized community vote-based recommendations and collaborative filtering recommendations to hybrid recommendations such as social network-based recommendation boosted by content information of items. We found that social network-based recommendations fused with content information and non-personalized community vote-based recommendations performed the best. Moreover, for cold-start users who have insufficient number of items to express their preferences, the recommendations based on their social connections generated significantly better predictions than other approaches.