Enriching very large ontologies using the WWW

  Request Code

  API Access   Call/Text an Expert

This paper explores the possibility to exploit text on the world wide web in order to enrich the concepts in existing ontologies. First, a method to retrieve documents from the WWW related to a concept is described. These document collections are used 1) to construct topic signatures (lists of topically related words) for each concept in WordNet, and 2) to build hierarchical clusters of the concepts (the word senses) that lexicalize a given word. The overall goal is to overcome two shortcomings of WordNet: the lack of topical links among concepts, and the proliferation of senses. Topic signatures are validated on a word sense disambiguation task with good results, which are improved when the hierarchical clusters are used.