A step towards high quality one-class Collaborative Filtering using online social relationships

Current available recommender systems mostly predict the target user's preferences using the traditional method, Collaborative Filtering (CF), which relies on people who share similar interests with the target user. Unfortunately, CF may lead to an invalid recommendation due to the lack of explicit feedback or item ratings from users in the real-world systems. One-class Collaborative Filtering (OCCF) became more realistic since it takes only positive examples or implicit feedback into consideration to provide better recommendations. The emergence of online social networking services which are growing at an explosive rate and the generate-and-share contents on countless number of news updates, opinions, interests and reviews lead to the social networking based recommendation approaches. Therefore, we wished to take advantage of social networking services to improve OCCF. In this paper, we propose a novel method for OCCF using online social relationships to increase a prediction accuracy of the recommendations. It is believed that social-relationship data can reflect the social influence, in other words, the interests of a user are similar to that of his/her friends in an online social network. Non-negative Matrix Factorization (NMF) method was applied with social influence weighting scheme to the one-class problem. Based on the experimental evaluation and two decision-support measures, our method presented proved to provide higher quality of recommendation results than the other baseline methods.