Confidence Management for Learning Ontologies from Dynamic Web Sources

Abstract

Dynamic environments require effective update mechanisms for ontologies to incorporate new knowledge. In this position paper we present a dynamic framework for ontology learning which integrates automated learning methods with rapid user feedback mechanism to build and extend lightweight domain ontologies at regular intervals. Automated methods collect evidence from a variety of heterogeneous sources and generate an ontology with spreading activation techniques, while crowdsourcing in the form of Games with a Purpose validates the new ontology elements. Special data structures support dynamic confidence management in regards to three major aspects of the ontology: (i) the incoming facts collected from evidence sources, (ii) the relations that constitute the extended ontology, and (iii) the observed quality of evidence sources. Based on these data structures we propose trend detection experiments to measure not only significant changes in the domain, but also in the conceptualization suggested by user feedback