An RDF Based Fuzzy Ontology Using Neural Tensor Networks

As an extension of classical ontology, a fuzzy ontology by employing fuzzy set theory can easily and yet better deal with uncertainties especially for the cases in which knowledge is vague. Obviously, fuzzification plays an important role in each fuzzy ontology. The main goal of this paper is to present an RDF based ontology, which indeed should contain many facts about the real world, inevitably facing with some uncertainties. In this perspective, an RDF based ontology is converted into a fuzzy most probably an incomplete one due to the fact that there will be some missing relations in the converted fuzzy ontology. To remedy this, the paper introduces a new method in the general framework of conversion and completion of an RDF based ontology into a fuzzy ontology mainly using the facts aspect. Therefore, first a new definition of the fuzzy ontology is proposed. To do so, a neural tensor network, which is indeed state-of-the-art of RDF based ontology completion, is proposed. Furthermore, a new application is suggested for this network that can create a fuzzy ontology. To furnish this goal, two new algorithms are then introduced for the conversion and completion of the proposed fuzzy ontology. In the proposed method, ontology facts are first embedded in a vector space, and then a score value is given to each fact by a learning method. Using these scores and threshold values of each relation, ontology facts can be fuzzified. Finally, some simulation studies are conducted to evaluate better the merit of the proposed method.