Learning Probabilistic Description Logics Theories

Uncertain information is ubiquitous in real world domains and in the Semantic Web. Recently, the problem of representing this uncertainty in description logics has received an increasing attention. In probabilistic Description Logics, knowledge bases contain numeric parameters that are often difficult to specify for a human. Moreover, the information are incomplete and poorly structured. On the other hand, data is usually available about the domain that can be leveraged for tuning the parameters and learn the structure of the information. In this paper we consider the problem of learning both the structure and the parameters of Probabilistic Description Logics under the DISPONTE semantics. We overview two systems we hve implemented: EDGE, that returns the value of the probabilities associated with axioms tuned using an Expectation Maximization algorithm, and LEAP, that exploits EDGE and the system CELOE to learn both the structure and the parameters of DISPONTE knowledge bases.