Mining DL ontologies
From International Center for Computational Logic
Mining DL ontologies
Talk by Ulrike Sattler
- Location: APB 3027
- Start: 15. June 2016 at 10:00 am
- End: 15. June 2016 at 11:00 am
- Research group: Automata Theory
- iCal
In a variety of domains and for a variety of purposes, people are using Description Logics (DL) ontologies: depending on how we count, there are hundreds or thousands of them. Quite often, these ontologies have a rather lightweight class hierarchy, and come with data in the form of an ABox or are used together in applications with external data. In the presence of this data, a naturally arising question is whether we can’t use this data to mine or learn general concept inclusion axioms, i.e., TBox axioms? And if yes, how can we do this, and how complex can these axioms be? In my talk, I will discuss these questions and describe our approach, which has been implemented in DL Miner. We aim at identifying hypotheses that capture interesting correlations in the data in the ontology. Our approach can be said to be logically and statistically sound: it works with DL’s open world assumption, and identifies suitable hypotheses by regarding this as a multi-dimensional optimisation problem (e.g., we are after hypotheses that are maximally supported and minimally brave). I will explain the setting for this problem, our approach, and our experiences with DL Miner. I will assume that the audience has a basic understanding of DLs, but this talk should also be of interest to anybody with some understanding of First Order Logic. This is joint work with Slava (Viachaslau) Sazonau.
