Use of social network information to enhance collaborative filtering performance

Abstract

When people make decisions, they usually rely on recommendations from friends and acquaintances. Although collaborative filtering (CF), the most popular recommendation technique, utilizes similar neighbors to generate recommendations, it does not distinguish friends in a neighborhood from strangers who have similar tastes. Because social networking Web sites now make it easy to gather social network information, a study about the use of social network information in making recommendations will probably produce productive results. In this study, we developed a way to increase recommendation effectiveness by incorporating social network information into CF. We collected data about users’ preference ratings and their social network relationships from a social networking Web site. Then, we evaluated CF performance with diverse neighbor groups combining groups of friends and nearest neighbors. Our results indicated that more accurate prediction algorithms can be produced by incorporating social network information into CF.