Advanced Self-Learning Ontologies

Traditional methods for developing ontologies involve a very time-intensive and error-prone process, which poses significant maintenance problems throughout the ontology lifetime. Furthermore, most users of ontologies treat them as static; which leads to a tendency for such systems to go for extended periods of time without any significant update. This is particularly true when the ontology describes a domain where the knowledge is perishable and being refreshed in near real-time, such as in many intelligence applications. SET has developed a new approach to effectively address the adaptive ontology challenge. During Phase 1, we developed a metamodel framework for ontologies that is domain independent. Through a robust proof-of-concept prototype, the SET team demonstrated its “Collective Learning Environment for Ontologies” (CLEO) architectural concept in Phase 1. SET proposes to extend this architecture and the associated implementation during Phase II to support minimally supervised learning of ontologies. The ultimate goal of this effort is to develop, refine, evolve, and maintain ontologies using a continuous stream of raw data, and to derive new structural and domain-specific information using machine learning algorithms. Together, these two goals will be met by a mixed-initiative, partially supervised learning system that allows the ontology to self-adapt to new information.