Automatic Population of a Pharmaceutical Company Ontology

We have implemented and evaluated a method to populate a company ontology, focusing on hierarchical relations such as acquisitions or subsidiaries. Our method searches for information about user-specified companies on the Internet using a search engine API (Google Custom Search API). From the resulted snippets we identify companies using machine learning and extract relations between them using a set of manually defined semantic patterns. We developed filtering methods both for companies and unlikely relations and from the set of company and relation instances we build this way, we construct an ontology addressing identity matching and consistency problems in a company-specific manner. We achieved a precision of 77 to 93 percent, depending on the evaluated relations.