A Proposal For Stable Semantic Metrics Based On Evolving Ontologies

The semantic web, which is called the next generation of Web, has been given extremely attention since it was proposed by Tim Berners-Lee. As the core of the semantic web, Ontology is all advanced technology of representing knowledge in Artificial Intelligence, where specified concepts and relationships are used to describe information. Ontology has been an important technology in Artificial Intelligence and knowledge engineering, and it is of great significance to the acquisition, representation, analysis and application of knowledge areas. So many research agencies in and abroad have done abundant research in this aspect.Assessing ontology quality has become an important issue because assessing ontology can help ontology engineers to predict the design quality of ontologies, and select ontologies, and even repair and improve the design of ontologies. Many ontology measures have been proposed and some principal work also has been done to study the nature of measures for ontologies in general. These metrics and principles provide useful guides about what measurement methods are considered and how useful the method is.Although these measures are also applicable to assessing ontology quality, they have encountered some problems. First, most metrics still are based on structural notions without indeed taking into account the semantics such as subsumption, which leads to incomparable measurement results. Second, most proposed metrics are not stable without considering possible additions of further axioms to ontology because they have not taken the open world assumption (OWA) into account, whereas OWA can indeed satisfy the requirements of ontologies in the context of dynamic and changing Web. Third, just because of changing and evolving characters of ontologies, consistent ontologies probably become inconsistent. But fewer metrics are considering measuring ontology inconsistency. In this paper, we propose a set of stable semantic cohesion metrics to assess the quality of evolving ontologies in the context of dynamic and changing Web. We argue that these metrics are stable and can be used to measure ontology semantics rather than ontology structures. These metrics are Average Axiom Fan outs per Class (AAFC), Number of Minimally Inconsistent subsets (NMIS) and Average Value of Axiom Inconsistencies (AVAI). The proposed semantic cohesion metrics are validated by using Kitchenham et al.'s and Briand et al.'s frameworks, and empirically validated by using a prototype implementing the metrics and algorithms presented in this paper.