A Rigorous Characterization of Classification Performance - A Tale of Four Reasoners.
Yong‐Bin Kang, Yuan-Fang Li, Shonali Priyadarsini Krishnaswamy · Swinburne figshare (Swinburne University of Technology) · 2012
A number of ontology reasoners have been developed for reasoning over highly expressive ontology languages such as OWL DL and OWL 2 DL. Such languages have, as a consequence of high expressivity, high worst-case complexity. Therefore, reasoning tasks such as classification sometimes take considerable time on large and complex ontologies. In this paper, we carry out a comprehensive comparative study to analyze classification performance of four widely-used reasoners, FaCT++, HermiT, Pellet and TrOWL, using a dataset of over 300 real-world ontologies. Our investigation on correlating reasoner performance with ontology metrics using machine learning techniques also provides additional insights into the hardness of individual ontologies.