Attach Topic Sense to Social Tags

Social tagging system, also noted as folksonomies, is an important way for users to describe resources on the Web. Because of the continually changing and informal definition, the semantics of these social tags are ambiguous and hard to adopt for web applications. In this paper, we propose a method to attach semantic topic sense to tags. The non-negative matrix factorization (NMF) is performed to find the hidden topics in the folksonomy. A novel automatic evaluation method is also proposed to measure our approach. Our evaluation shows that the topic sense induction in a folksonomy allows for precise and complete search, which is one of the key functionalities in social tagging systems.