対話型進化計算による感性検索エージェントの最適化

抄録

This paper proposes a method of optimizing Kansei retrieval agents for multimedia data retrieval system. The system has multiple agents, each of which has a different character, to retrieve various multimedia data. The agent has a Kansei model controlled by Kansei parameters and can change its emotional response by changing values of the parameters. The proposed system can retrieve data desired by the user if the agents are optimized to adapt them to the user's particular emotional responses. In the proposed method, the system uses Interactive Genetic Algorithms (IGA) to optimize the agents. Evaluation of each agent in IGA population is determined by the user's evaluation to multiple data retrieved by multiple agents. This paper proposes two advanced evolution methods for optimizing Kansei retrieval agents in IGA operation, which exhibit better performances in evolving Kansei models compared to previous method.