Class Extraction from the World Wide Web

In previous work we introduced KNOWITALL a data-driven, Web-based information extraction system. This paper focuses on the task of automatically extending the system’s initial ontology by extracting subclasses of given general classes and by discovering other closely related classes. We first show that the basic KNOWITALL model can be easily extended to accommodate high-precision subclass extraction (SE) and that subclass knowledge can lead to increases by a factor of 10 in the number of class instances extracted by KNOWITALL. Second, we introduce two methods that on average increase the number of subclasses extracted by the SE module itself by a factor of 3. Next, we show how the model can be extended to achieve high-quality related class extraction (RCE), which consists of discovering new classes closely related to the set of known classes. Of the classes discovered by KNOWITALL, in the test domains of Geography and Computers, 79.5% were bona-fide related classes.