Efficient estimation of ontology entities distributed representations

Ontologies have been used as a form of knowledge representation in different fields such as artificial intelligence, semantic web and natural language processing. The success caused by deep learning in recent years as a major upheaval in the field of artificial intelligence depends greatly on the data representation, since these representations can encode different types of hidden syntactic and semantic relationships in data, making their use very common in data science tasks. Ontologies do not escape this trend, applying deep learning techniques in the ontology-engineering field has heightened the need to learn and generate representations of the ontological data, which will allow ontologies to be exploited by such models and algorithms and thus automatizing different ontology-engineering tasks. This paper presents a novel approach for learning low dimensional continuous feature representations for ontology entities based on the semantic embedded in ontologies, using a multi-input feed-forward neural network trained using noise contrastive estimation technique. Semantically similar ontology entities will have relatively close corresponding representations in the projection space. Thus, the relationships between the ontology entities representations mirrors exactly the semantic relations between the corresponding entities in the source ontology.