Information Extraction for Ontology Learning

In this chapter, we discuss how ontologies can be constructed by extracting information from Web documents. This is a challenging task, because information extraction is usually a noisy endeavor, whereas ontologies usually require clean and crisp data. This means that the extracted information has to be cleaned, disambiguated, and made logically consistent to some degree. We will discuss three approaches that extract an ontology in this spirit from Wikipedia (DBpedia, YAGO, and KOG). We will also present approaches that aim to extract an ontology from natural language documents or, by extension, from the entire Web (OntoUSP, NELL and SOFIE). We will show that information extraction and ontology construction can enter into a fruitful reinforcement loop, where more extracted information leads to a larger ontology, and a larger ontology helps extracting more information.