Construction of a Local Domain Ontology from News Stories

The identification of "actionable" information in news stories has become a popular area for investigation. News presents some unique challenges for the researcher. The size constraints of a news story often require that full background information is omitted. Although this is acceptable for a human reader, it makes any form of automatic analysis difficult. Computational analysis may require some background information to provide context to news stories. There have been some attempts to identify and store background information. These approaches have tended to use an ontology to represent relationships and concepts present in the background information. The current methods of creating and populating ontologies with background information for news analysis were unsuitable for our future needs. In this paper we present an automatic construction and population method of a domain ontology. This method produces an ontology which has the coverage of a manually created ontology and the ease of construction of the semi-automatic method. The proposed method uses a recursive algorithm which identifies relevant news stories from a corpus. For each story the algorithm tries to locate further related stories and background information. The proposed method also describes a pruning procedure which removes extraneous information from the ontology. Finally, the proposed method describes a procedure for adapting the ontology over time in response to changes in the monitored domain.