Exposing Latent Information in Folksonomies for Reasoning

User tagging by means of authority-free folksonomies has become standard for making online videos, photos, bookmarks and blog posts discoverable. We propose to develop a system that identifies and exposes the ontological structure that is latent in the folksonomic tags used in open, publicly available sites such as Flickr (photos), YouTube (video) and Blogger (blogs) and demonstrates the ability to perform higher order processing using the induced structure. Specifically, the technology developed will allow the system to identify photos, videos or blog posts about the same event by means of the tags applied in one medium (e.g. photos) in order to identify tagged items in another medium (videos or blog posts). We plan to use (i) the surrounding text, (ii) the user’s other tagging practices, and, most importantly, (iii) large-scale topic hierarchies used in DMOZ, Wikipedia (dbPedia) and lexical resources such as Beth Levin’s taxonomy of verb classes in order to tokenize and disambiguate the tags into unambiguous topics. These will then be mapped onto who, what, when, and where slots, and used to identify depictions of the same event in other media using a formal reasoner, VIStology’s BaseVISor, a forward-chaining OWL/RDF reasoner using r-entailment.