Home // International Journal On Advances in Software, volume 12, numbers 3 and 4, 2019 // View article

Creating and Evaluating Data-Driven Ontologies

Authors:
Maaike H.T. de Boer
Jack P.C. Verhoosel

Keywords: Ontologies; Machine Learning; NLP; Word2vec; Ontology Learning; F1 score

Abstract:
Automatically creating data-driven ontologies is a challenge but it can save time and resources. In this paper, eight data-driven algorithms are compared to create ontologies, four ontologies based on documents and four based on keywords, on three different document sets. We evaluate the performance using three different evaluation metrics based on nodes, weights and relations. Results show that 1) keyword-based methods are in general better than document-based methods; 2) a co-occurrence-based algorithm is the best document-based method; 3) the evaluation metrics give useful insight, but need to be enhanced in future work. It is advised to a) use the created ontologies as a head start in an ontology creation session, but not use the ontologies as created; b) use word2vec to generate an ontology in a generic domain, whereas the co-occurrences algorithm should be used in specific domain.

Pages: 300 to 309

Copyright: Copyright (c) to authors, 2019. Used with permission.

Publication date: December 30, 2019

Published in: journal

ISSN: 1942-2628