Dynamic Integration of Multiple Evidence Sources for Ontology Learning

Abstract

Although ontologies are central to the Semantic Web, current ontology learning methods primarily make use of a single evidence source and are agnostic in their internal representations to the evolution of ontology knowledge. This article presents a continuous ontology learning framework that overcomes these shortcomings by integrating evidence from multiple, heterogeneous sources (unstructured, structured, social) in a consistent model, and by providing mechanisms for the fine-grained tracing of the evolution of domain ontologies. The presented framework supports a tight integration of human and machine computation. Crowdsourcing in the tradition of games with a purpose performs the evaluation of the learned ontologies and facilitates the automatic optimization of learning algorithms.