Exposing commercial value in social networks: matching online communities and businesses

This paper explores the problem of user recommendations in group settings. Research in networked social behavior has found evidence that individuals tend to form connections to those having similar interests to their own. Known as homophily, this tendency has been shown to significantly inform the structure of social networks. Collaborative filtering (CF) techniques, however, typically do not take into account explicit social relationships among users in predicting user preferences, although the importance of peer influence in user consumption and marketing has long been recognized. We test for the existence of homophily on the Yelp! dataset and find ample evidence that social ties are indicative of similar preferences. Based on this insight, experimental results show we are able to improve on traditional CF methods by 1) using clustering algorithms to detect communities of users with similar tastes, and 2) incorporating information about social structure in CF techniques. We find that leveraging social structure information improves the performance of CF and reveals relevant and valuable matches between businesses and the members of the group. These results have important implications for targeted marketing.