Attribute Extraction from Conjectural Queries
MARIUS A. PAŞCA · 2012
Conjectural search queries (is python case sensitive, is millennium stadium heated) embody attempts by Web users to verify whether a particular property (soluble in water?, case sensitive?, heated?) does or does not apply to a particular instance (iodine, python, millennium stadium). This paper considers such queries to be a data source of attributes of open-domain classes. Conjectural attributes complement attributes encoded in human-compiled knowledge resources or automatically acquired from text by previous methods. They correspond to properties of interest to Web users, which are not necessarily stated in nominal form. Relevant properties of Chemical elements, Programming languages and Stadiums include whether they are soluble in water, flammable or ductile; case sensitive, platform independent, or interpreted; or air conditioned, roof retractable or heated, respectively. Experimental results show that relevant, conjectural attributes can be extracted from inherently-noisy queries, for a variety of open-domain classes of interest.