Foundations of Onto-Relational Learning

ILP is a major approach to Relational Learning that exploits previous results in concept learning and is characterized by the use of prior conceptual knowledge. An increasing amount of conceptual knowledge is being made available in the form of ontologies, mainly formalized with Description Logics (DLs). In this paper we consider the problem of learning rules from observations that combine relational data and ontologies, and identify the ingredients of an ILP solution to it. Our proposal relies on the expressive and deductive power of the KR framework +log that allows for the tight integration of DLs and disjunctive Datalog with negation. More precisely we adopt an instantiation of this framework which integrates the DL and positive Datalog. We claim that this proposal lays the foundations of an extension of Relational Learning, called Onto-Relational Learning, to account for ontologies.