The Research For Image Tag Disambiguation In Collaborative Annotation System

With the development of social network, multimedia technology and communication services, the popularity of digital equipment, and the massive growth of digital video/photo, researches on social media are becoming a hot spot. Network resource retrieval is mainly carried out by absolute keyword matching. Tag is one of the most commonly used text-matching objects for retrieval.Most users in collaborative annotation system are not professionals, therefor many tags are assigned with lots of randomness and diversity. A tag can have several meanings, which lead to the ambiguity of the labeled object’s true content. In order to improve the accuracy of the tags, polysemy and synonymy are the most important problems need to be solved.In this paper, we designed a context-based tag disambiguation algorithm. First we expand the tag semantic by making use of Wikipedia’s knowledge and get several candidate concepts. Then extract the image context information and compare it with each candidate concept. Finally we choose the most relevant concept as the true meaning of the tag.