A Practical Ontology-Based Concept Learning in MAS

Agent-mediated knowledge management is believed to be an efficient solution for knowledge sharing among different software systems. In this paradigm, diverse systems can exchange information while keeping their own individual ontology to represent their domain knowledge. However, overcoming semantic heterogeneity among diverse ontologies becomes a key issue. We propose a new concept learning mechanism to solve this problem and also implement it using multiagent technology together with IBM's UIMA (Unstructured Information Management Architecture) into a semantic search application. In this method an agent in one system can learn a concept which it does not know by getting advices from agents in other systems and integrate the learnt concept into its local ontology. In addition, we successfully integrate our concept-learning mechanism with the semantic search application. Through the semantic interoperation between concept learning and semantic search, the qualities of both concept learning and semantic search are largely increased.