Results

In the scatter plot of figure fsr we can identify two regions with distinctly different characteristics: The concept pairs of the relatively small vehicle ontology are represented with blue dots and are in the range of 4 to 8 features the concept pairs of the species ontology contain a total of 10 to 20 different features. Here are a number of lessons to be learned from analysis.

• Since vehicle and species have not yet shared any characteristics, the concept pair V/SP has the only FSR value at 0% that is below 50%. This also corresponds to the expectation, because intuition would also immediately say that one has nothing to do with the other.
• Pairs of concepts relatively high up in the class hierarchy like (T=^Truck, P=^Pickup), (SP=^Species, STP=^Species_by_Threat-Phase), (STP=^Species_by_Threat-Phase, ^Bird ) etc., also have a high FSR value, usually well above 75%, as expected.
• The area where the sum of the features of the concept pairs is between 13 and 20 is conspicuous in the species. Here the FSR values are in a range of approx. 60% to 75%. Not surprisingly, concept bars containing either birds or bears have relatively close stats.
• On the other hand, the relatively high similarity of 58.8% between emperor penguins (EP) and polar bears (PLB) is surprising, although both are completely different species. However, if one looks more closely, one notices that they share many secondary characteristics, derived primarily from the definition of concepts via genus differentiae patterns. Both species live in a pole region, are piscivores and their habitat is maritime.
• For the vehicles we have five concept pairs with an average of 5.4 features and an average feature similarity ratio (FSR) of 70%, for the species there are 10 concept pairs with an average of 16 features and an average feature similarity ratio of 74%.
• The maximum FSR value of 100% can only be reached if two direct subclasses of a class have no features. In practice, this should make little sense because the subclasses should have at least one differentiating feature. This is e.g. the case in the original modeling of [GuAl2015] bird and bird species (see section Variable Embodiment. There all pairs of classes do have an FSR value of 100%. The remodeling carried out with our methods in section Correctness first gives rise to the possibilities of a more differentiated view. This provides another important argument for using the methods we have presented for more expressive ontology modelling.

With the procedure for measuring feature similarities, we have for the first time developed a method that makes the similarity of concepts in ontologies measurable. As shown, valuable conclusions can be drawn about the relatedness of concepts and the correctness of ontologies.