Abstract
This paper describes a query classification system for a specialized domain. We take as a case study queries asked to a search engine of an art, cultural and history library and classify them against the library cataloguing categories. We show how click-through links, i.e., the links that a user clicks after submitting a query, can be exploited for extracting information useful to enrich the query as well as for creating the training set for a machine learning based classifier. Moreover, we show how Topic Model can be exploited to further enrich the query with hidden topics induced from the library meta-data. The experimental evaluations show that this system considerably outperforms a matching and ranking classification approach, where queries (and categories) were also enriched with similar information.