Domain-Specific Query Augmentation using Folksonomy Tags: the Case of Contextual Advertising

Folksonomies allow users to collaboratively tag a variety of tex- tual and multimedia objects with sets of labels. The largest folk- sonomy projects, such as FLICKR and DEL.ICIO.US, contain mil- lions of multi-labeled objects, and embed significant amounts of human knowledge. We propose a method for automatically using this knowledge to augment traditional IR systems, using contextual advertising as an application domain. Given a query, we first iden- tify a set of relevant tags, and then use tags that cooccur with them to augment the query. Importantly, our method performs domain- specific query disambiguation, and can actually learn that a query "menu" is likely to have food connotation on FLICKR but user in- terface connotation on DEL.ICIO.US.