Improving Example Selection for Agents Teaching Ontology Concepts

We present a method to improve the positive examples selection by teaching agents in a multi-agent system in which a team of agent peers teach concepts to a learning agent. The basic idea in this method is to let a teacher agent expand the features it uses to describe a concept in its ontology by additional features. This resembles the typical behavior of human teachers who describe concepts from different viewpoints in the hope that one of these viewpoints comes close to the viewpoint of a learner. The extended feature set is then used to select positive examples that together with negative examples are communicated to the learner agent. The learner uses concept learning techniques to integrate the new concept into its own ontology. An experimental evaluation shows a significant learning improvement compared to the previous approach.