Bottom-up enrichment of top-down ontologies through annotation

Description

We present a methodology to enhance domain-specific ontologies by (i) addressing a manual annotation of texts with the concepts in the domain ontology, (ii) matching the annotated concepts with the closest YAGO-Wikipedia concept available, and (iii) using concepts from other ontologies that cover complementary domains. This method reduces the difficulty of aligning ontologies, because annotators are asked to associate two labels from different inventories to a concrete example, which requires a simple judgment. In a second phase, those correspondences are consolidated into a proper alignment. The resulting alignment is a partial connection between diverse ontologies, and also a strong connection to Linked Open Data. By aligning these ontologies, we are increasing the ontological coverage for texts in that domain. Moreover, by aligning domain on-tologies to the Wikipedia (via YAGO), we can obtain manually annotated examples for some of the concepts, effectively populating the ontology with examples. We present two applications of this process in the legal domain. First, we annotate sentences of the European Court of Human Rights with the LKIF ontology, at the same time matching them with the YAGO ontology. Second, we annotate a corpus of customer questions and answers from an insurance web page with the P&C ontology for the insurance domain, matching it with the YAGO ontology and complementing it with a financial ontology.

Abstract

International audience

Additional details