Tag Clustering with Social Bookmarking Data

We propose a tag clustering with social bookmarking data in this paper. In social bookmarking services, various kinds of data, for example, text data, image data, and movie data, and users use tags, which are keywords added to registered web pages, to administer the registered web pages. However, the tags are not limited to a particular vocabulary and are added to the web pages freely. Hence, though the same tag is added to web pages, it might have a difference of meaning for each web page. Thesauruses are not useful to solve it because of including neologisms and symbols. Hence, our goal is to solve the ambiguity of tags and to classify them according to their meanings. To achieve the goal we regard adding a tag to a web page as a link between the tag and the web page and construct a weighted bipartite graph between tags and web pages without their contents. To classify the tags according to their meanings, Probabilistic Latent Semantic Indexing is used to analyze the weighted bipartite graph. We carried out evaluation experiments using real social bookmarking data, Buzzurl and confirmed the proposed method classifies the tags precisely regardless of the ambiguity of description and meaning.