Semantically-based Recommendation by using Semantic Clusters of Users’ Viewing History

Abstract

Collaborative filtering-based recommender systems assume that if two users have shown similar interest on the same set of contents, they may show a similar interest-pattern in choosing future contents. However, users who have certain tastes on one specific category of contents may behave differently in choosing contents from other categories. Moreover, the collaborative filtering approaches may not work efficiently with sparse data sets, where there are a small number of contents or a limited number of users in the content categories. In this paper, to deal with these problems, we propose a novel approach of recommending contents across different categories by considering both the semantic information of contents and user interests. The proposed approach uses Linked Data as the source to find the appropriate semantics of the contents extracted from users' viewing history. The semantic concepts retrieved for the contents are then grouped together into semantic clusters based on their similarity and relevance. Our approach recommends potentially interesting contents to general users based on the content-consumption trends monitored from leading user groups who most proactively and frequently consume contents. We demonstrate the effectiveness of our approach by conducting an experiment with the real-world mobile IPTV service data obtained from one of the popular IPTV service providers in South Korea.