Predicting the Possibilistic Score of OWL Axioms through Modified Support Vector Clustering

We address the problem of predicting a score for candidate axioms within the context of ontology learning. The prediction is based on a learning procedure based on support vector clustering originally developed for inferring the membership functions of fuzzy sets, and on a similarity measure for subsumption axioms based on semantic considerations and reminiscent of the Jaccard index. We show that the proposed method successfully learns the possibilistic score in a knowledge base consisting of pairs of candidate OWL axioms, meanwhile highlighting that a small subset of the considered axioms turns out harder to learn than the remainder.