Alignment Techniques in Domain-Specific Models

Abstract

During the requirements specification stage, conceptual descriptions of the Domain Universe (DU) are used through several representations such as glossaries, ontologies and other models. However, a complete representation of a domain is difficult to achieve. Different ways a domain can be conceptualized result in a heterogeneous representation, which often makes consistent description challenging. This work aims to identify techniques used in different projects for domain alignment and their behaviour with a lexical model. In the applied methodology, a search for the domain alignment was carried out in various scientific databases such as IEEE Explore, ACM and Springer. The obtained result showed that tools used to align domain specifications are concentrated in ontologies and glossaries, using techniques with similarities in the semantic, lexical, structural and syntactic fields. When applying semantic and lexical level measures in glossaries called Extended Language Lexicon (LEL) it was obtained the degree of similarity between the different glossaries, at the semantic level the one with the best precision was the Cosine Similarity method, while at the lexical level, the one with the best precision was Jaro Winkler. It is concluded that the application of the techniques independently does not allow the identification of similarities, omissions or correspondences between the domains, so it is necessary to develop a method that contemplates heuristics and integrates the different lexical and semantic measures as a support to improve the completeness and consistency of the domain.