Performance Evaluation of a Kansei Retrieval Agent Model with Neural Networks

We describe the effectiveness of a Kansei Retrieval Agent (KRA) model with Neural Network (NN) for retrieving user preference data. Our system evaluates KRAs using the user's evaluation and optimizes the KRA parameters using an evolutionary computation approach. While our previous study involved Kansei retrieval systems for T-shirts or music, the basic optimization performance of a KRA with NN was not examined. Therefore, in the present study, we investigate the effectiveness of the KRA model with an NN using numerical simulations. The simulation uses a pseudo user that imitates user preferences. We examine the quantization of evaluation values by the user and KRAs and compares the performances obtained using the particle swarm optimization and genetic algorithm methods. The simulation results show that the KRA model can learn user preferences.