PN-OWL: A two-stage algorithm to learn fuzzy concept inclusions from OWL 2 ontologies

Given a target class T of an OWL 2 ontology, positive (and possibly negative) examples of T, we address the problem of learning, viz. inducing, from the examples, fuzzy class inclusion rules that aim to describe conditions for being an individual classified as an instance of the class T.

To do so, we present PN-OWL which is a two-stage learning algorithm consisting of a P-stage and an N-stage. In the P-stage, the algorithm learns fuzzy class inclusion rules (the P-rules). These rules aim to cover as many positive examples as possible, increasing recall, without compromising too much precision. In the N-stage, the algorithm learns fuzzy class inclusion rules (the N-rules), that try to rule out as many false positives, covered by the rules learnt at the P-stage, as possible. Roughly, the P-rules tell why an individual should be classified as an instance of T, while the N-rules tell why it should not.

PN-OWL then aggregates the P-rules and the N-rules by combining them via an aggregation function to allow for a final decision on whether an individual is an instance of T or not.

We also illustrate the effectiveness of PN-OWL through extensive experimentation.