Arabic ontology extraction model from unstructured text

Abstract

Ontologies enable the sharing and reusing of knowledge, allowing the interoperation and combination of information systems. Automatic ontological relation extraction from text is an important factor for representing documents and their contents in a useful computerized manner. This paper proposes a model for automatic ontology extraction from Arabic text by parsing sentences and extracting part of speech (POS). Then, the proposed rule-based model is applied to extract the triple attributes of a sentence (subject, predicate, and object) from the parsing tree. Finally, the semantic relations can be used to extract new triples inferred from the initially extracted triples. The results were evaluated using two methods: the first method was based on a comparison with manual extraction because there is no standard method for measuring Arabic triple extraction; the second method is based on translating the dataset into English and comparing the output results using the Stanford dependencies extraction web tool. The proposed model achieved an accuracy of 73.6% for Arabic triple extraction and a 35% increase in overall triples owing to new inferred triples. However, when the dataset was translated into English and tested by the Stanford dependencies extraction web tool, the accuracy was 71.8% without inferred triples.