Enhancing the open-domain classification of named entity using linked open data

Abstract

Many applications make use of named entity classification. Machine learning is the preferred technique adopted for many named entity classification methods where the choice of features is critical to final performance. Existing approaches explore only the features derived from the characteristic of the named entity itself or its linguistic context. With the development of the Semantic Web, a large number of data sources are published and connected across the Web as Linked Open Data (LOD). LOD provides rich a priori knowledge about entity type information, knowledge that can be a valuable asset when used in connection with named entity classification. In this paper, we explore the use of LOD to enhance named entity classification. Our method extracts information from LOD and builds a type knowledge base which is used to score a (named entity string, type) pair. This score is then injected as one or more features into the existing classifier in order to improve its performance. We conducted a thorough experimental study and report the results, which confirm the effectiveness of our proposed method. © 2010 Springer-Verlag.