Exploiting tags for concept extraction and information integration

The use of tags to annotate content creates an opportunity to explore alternatives to automate the process of extracting semantics from data sources. Semantic information is needed for many complex tasks like concept extraction and information integration. In order to establish the value of user-generated annotation, this paper presents two experiments on which only user tags are used as input. At the core of semantic extraction is the identification of concepts and relationships that are present in the data. We show, through an experimental study on tagged photographs, how to extract concepts associated with photographs and their relationships. Our experiments demonstrate that supervised machine learning techniques can be used to extract a concept associated with a photograph with an overall precision score of 80%. Our experiments also show that a variation of the Jaccard similarity coefficient on sets of tags can be used to determine equivalence relationships between the concepts associated with these sets.