Creating ontologies for content representation—the OntoSeed suite

Abstract. Due to the inherent difficulties associated with manual ontology building, knowledge acquisition approaches such as ontology reuse or ontology learning from texts are often seen as instruments that can make this tedious process easier. In this paper we present a NLP-based method to aid ontology design in a specific application scenario, namely that in which the resulting ontology is used to support the semantic annotation of text documents. The proposed method uses the World Wide Web in its analysis of the domain-specific documents, thereby greatly reducing the need for linguistic expertise and resources, and suggests ways to specify domain ontologies in a “linguistics-friendly ” format in order to improve further ontology-based natural language processing tasks such as semantic annotation. We present a thorough evaluation of the method, using corpora from three diverse real-world settings (medical information, tourism, and recipes). Additionally, for the first scenario we compare the costs and the benefits of the NLP-based ontology engineering approach against a similar, reuse-oriented experiment.