Adjudicator: A Statistical Approach for Learning Ontology Concepts from Peer Agents

— We present a statistical approach for software agents to learn ontology concepts from peer agents by asking them whether they can reach consensus on significant differences between similar concepts. This method allows agents that are not sharing common ontologies to establish common grounds on concepts known only to some of them, when these common grounds are needed. The method starts with fetching positive and negative examples for a concept vaguely understood by a learner agent from the peer agents. The learner agent then uses a concept learning method to learn the concept in question. Then example objects of the candidate concept are sent back to the peer agents asking for their feedback. Peer agents evaluate the examples using two dimensional rate and weight evaluation criteria. The returned data is then tested for integrity and analyzed using analysis of variance to identify whether a statistical consensus can be achieved among peer agent with respect to the learnt concept. If...

• Type: Conference
• Year: 2007
• Where: SEKE
• Authors: Behrouz Homayoun Far, Abdel Halim Elamy, Nora Houari, Mohsen Afsharchi
• Keywords: Learner Agent, Peer Agent, SEKE 2007, Software Agents