Learning by googling

Abstract

The goal of giving a well-defined meaning to information is currently shared by endeavors such as the Semantic Web as well as by current trends within Knowledge Management. They all depend on the large-scale formalization of knowledge and on the availability of formal metadata about information resources. However, the question how to provide the necessary formal metadata in an effective and efficient way is still not solved to a satisfactory extent. Certainly, the most effective way to provide such metadata as well as formalized knowledge is to let humans encode them directly into the system, but this is neither efficient nor feasible. Furthermore, as current social studies show, individual knowledge is often less powerful than the collective knowledge of a certain community.As a potential way out of the knowledge acquisition bottleneck , we present a novel methodology that acquires collective knowledge from the World Wide Web using the Google TM API. In particular, we present PANKOW, a concrete instantiation of this methodology which is evaluated in two experiments: one with the aim of classifying novel instances with regard to an existing ontology and one with the aim of learning sub-/superconcept relations.