Diversified Coverage based Tag Recommendation

Tag recommendation, as a branch of recommendation engine, has drawn more and more attention, which is also extensively exploited in e-commerce and SNS (Social Networking Services). The results generated by the current algorithms could describe the items with a high relevance. However, they are often of poor diversity in the recommended results. That indicates there is a redundancy in the results in term of semantics. Such a case reduces the novelty and diversity of the recommended results, seriously affecting the user’s experience. In this paper, we define the tag correlation metric based on the local and global tag co-occurrence matrices, which improves the recommendation accuracy by incorporating both the user’s interests and the popularity of tags. Moreover, we propose the concept of semantic coverage, by which the redundancy of semantics can be removed efficiently. To our best knowledge, it is first proposed in the context of tag recommendation. Finally, a diversified coverage based tag recommendation algorithm, namely EDC, is developed. By converting the problem of diversified coverage tag recommendation to the MIDS (Minimum Independent Dominating Set) problem, EDC first handles the cliques and the bipartites in the graph. Then, it recursively searches the MIDSs in the remaining graph. Further, a greedy algorithm GDC is proposed. The experiments conducted on the real datasets of MovieLens and Last.fm show that the proposed EDC and GDC improve the diversity significantly.