Using Tags for Measuring the Semantic Similarity of Users to Enhance Collaborative Filtering Recommender Systems

Recent years have seen a significant growth in social tagging systems, which allow users to use their own generated tags to organize, categorize, describe and search digital content on social media. The growing popularity of tagging systems is leading to an increasing need for automatic generation of recommended items for users. Much previous research focuses on incorporating recommender techniques in social tagging systems to support the suggestion of suitable tags for annotating related items. Collaborative filtering is one such technique. The most critical task in collaborative filtering is finding related users with similar preferences, i.e., “liked-minded” users. Despite the popularity of collaborative filtering, it still suffers from certain limitations in relation to “cold-start” users, for example, where often there are insufficient preferences to make recommendations. Moreover, there is the data-sparsity problem, where there is limited user feedback data to identify similarities in users’ interests because there is no intersection between users’ transactional data a situation which also results in degraded recommendation quality. For this reason, in this paper we present a new collaborative filtering approach based on users’ semantic tags, which calculates the similarity between users by discovering the semantic spaces in their posted tags. We believe that this approach better reflects the semantic similarity between users according to their tagging perspectives and consequently improves recommendations through the identification of semantically related items for each user. Our experiment on a real-life dataset shows that the proposed approach outperforms the traditional user-based collaborative filtering approach in terms of improving the quality of recommendations.