Tag propagation based recommendation across diverse social media

Abstract

Many real applications demand accurate cross-domain recommendation, e.g., recommending a Weibo (the largest Chinese Twitter) user with the products in an e-commerce Web site. Since many social media have rich tags on both items or users, tag-based profiling became popular for recommendation. However, most previous recommendation approaches have low effectiveness in handling sparse data or matching tags from different social media. Addressing these problems, we first propose an optimized local tag propagation algorithm to generate tags for profiling Weibo users and then use a Chinese knowledge graph accompanied by an improved ESA (explicit semantic analysis) for semantic matching of cross-domain tags. Empirical comparisons to the state-of-the-art approaches justify the efficiency and effectiveness of our approaches.