Topic sense induction from social tags based on non-negative matrix factorization

Social tagging, also noted as collaborative tagging or folksonomy, is an important way for users themselves to describe resources on the Web. The tags that the web users adopt to describe the resources are called social tags, and they have been widely used and studied. However, for the absence of a central controlled vocabulary, the semantics of the social tags are ambiguous due to constant changes of either the users’ interests or the informal definitions, which makes it hard to directly make use of these social tags in the web applications. In this paper, we propose a non-negative matrix factorization (NMF) based method to automatically induce topic senses from social tags, which can then be used for the tag disambiguation. A novel automatic evaluation method is also proposed to evaluate our method. The experiment results show that the proposed topic sense induction method can help to provide precise resources search and recommendation, which is one of the key functionalities in social tagging systems.