The Role of Activity and Similarity in Rating and Social Behavior in Social Recommender Systems

Abstract

Listen

Translate article

Social recommender systems exploit two sources of information for making recommendations, the historical rating behavior of users, and the social connections among them. The basic assumption is that if two users are friends, they are likely to share similar preferences. Many recommendation approaches are based on such correlations between the rating and the social behavior of users. However, there is little work in studying whether there actually exist such correlations and how strong they are. In our work, we look at the two views of user behavior, their social connections, and their history of ratings, and investigate two research questions. The first examines if strong activity in one view, e.g., having many friends, implies strong activity in the other view, e.g., having rated many items. The second investigates whether high similarity in one view, e.g., network similarity, implies high similarity in the other view, e.g., rating similarity. We employ various notions of activity and similarity, and identify those that appear to have the stronger impact. Specifically, to some degree, we find that rating behavior determines social behavior, and that the opposite relationship is weaker.