Ontology based entity disambiguation with natural language patterns

Abstract

Text is still the most important carrier of information. Names are essential anchor points in text, and they are essential to link pieces of information spread in the text. Unfortunately names are ambiguous, which makes it difficult for systems to understand whether syntactically identical names are actually semantically equivalent. Disambiguation is therefore an indispensable feature for the automatic processing of texts. We introduce a novel approach to dis-ambiguate names using ontologies, natural language text patterns and a customisable ranking algorithm. We evaluate our approach in the area of geographic names, and demonstrate the utility of our results with a geographic news reader.