Ontology Reuse: Neural Network-Based Measurement of Concepts Representations and Similarities in Ontology Corpus

Ontologies are the heart of the semantic web. They are designed to be reused in web applications. This paper aims to discover a given concept's representation against existing ontologies in a corpus, and if the concept is represented, other similar concepts and terms to it are extracted. A corpus formed of several ontologies in the agricultural domain was constructed. SPARQL queries were used to extract the required data from existing ontologies. And a machine learning technique, the Word2Vec, was employed for ontology reuse process to measure concepts similarity against the existing ontologies. The experimental results showed that the proposed methodology successfully detected previously seen vocabularies during the training on the data in the ontology corpus, and retrieved other similar concepts from the ontologies as well as their degree of similarity (Cosin similarity). Furthermore, the proposed model could process over two million terms in around one minute, reflecting its effectiveness in this context. The proposed method would be useful to ontology and knowledge engineers to conduct a preliminary investigation about which existing ontologies are suitable for reuse in the process of developing new ontologies. Other applications of the proposed method may include ontology alignment to measure the degree of similarity between existing ontologies.