Advanced Social Recommendations with SoNARS++

Recommender systems support users in finding the appropriate information at the correct time. While traditional recommenders only take into account the stable preferences of target users (ego-based interests), some recent approaches in the area of social recommender systems have started to acknowledge that the mere fact of taking part in social relationships may cause individuals to modify their attitudes and behaviors, and have proposed methods for generating recommendations based on the preferences of the target users' social networks (network-based interests). However, little work has investigated how to effectively merge ego- and network-based interests. In this paper, we present SoNARS++, an advanced social algorithm that assesses the interest for an item to be recommended by combining a user's personal interests for that item with the interests for that item of the user's social network, depending on its structure and on social influence relationships among users. The results of the experimental evaluation we carried out show that SoNARS++ is comparable for its precision to a mainstream algorithm such as collaborative filtering but appears to provide users with more useful recommendations.