Recommending Talks at Research Conferences Using Users' Social Networks

Abstract

This paper investigates recommendation algorithms to suggest talks of interest to attendees of research conferences. In this study, based on a social conference support system Conference Navigator 3 (CN3), we explored three kinds of knowledge sources to generate recommendations: users' preference about talks (CN3 bookmarks), users' social networks (research collaboration network and CN3 following network) and talk content information (titles and abstracts). Using these sources, we explored a diverse set of algorithms from non-personalized community vote-based recommendations and conventional collaborative filtering recommendations to hybrid recommendations such as social network-based (SN) recommendations boosted by content information of talks. We found that SN recommendations fused with content information outperformed the other approaches. Moreover, for cold-start users who have an insufficient number of bookmarks to express their preferences, the recommendations based on their social connections also generated significantly better suggestions than the other approaches. Between two kinds of social networks that we considered as foundations of recommendations, there was no significant difference in the quality of the recommendations.