Incorporating Social Actions into Recommender Systems

Abstract

With the rapidly growing amount of information available on the internet, recommender systems become popular tools to promote relevant online information to a given user. Although collaborative filtering is the most popular approach to build recommender systems and has been widely deployed in many applications, it still pay little attention to social actions, which are widely common in social networks and we believe could make a significant improvement in recommender systems. In this paper, we incorporate users’ social actions into a model-based approach for recommendation using probabilistic matrix factorization. Compared with previous work, users’ social actions are taken as a new relation to optimize previous trust-based recommender systems. To achieve this, we propose a social recommendation graphical model employing users’ relations based on their social actions. We make use of users’ commenting action in our approach and conduct experiments on a real life dataset, extracted from the Douban movie ratings and comments system. Our experiments demonstrate that incorporating users’ social action information leads to a significant improvement in recommender systems.KeywordsRecommendationSocial NetworkSocial ActionsProbabilistic Matrix Factorization