Interactive genetic algorithm with brain activation measured by functional magnetic resonance imaging

• Published in: 2015 IEEE Congress on Evolutionary Computation (CEC) pp. 2222 - 2229
• Main Authors: Tanaka, Misato, Miki, Mitsunori, Yamamoto, Utako, Hiroyasu, Tomoyuki
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2015
• Subjects: Blood; Brain; Estimation; Magnetic resonance imaging; Neuroimaging; Optimization; Support vector machines
• ISSN: 1089-778X; 1941-0026
• DOI: 10.1109/CEC.2015.7257159

The purpose of this study is to develop a product recommendation system using interactive optimization based on human Kansei. Introduction of brain information to Interactive Genetic Algorithm (IGA) was considered, because this biological information is strongly related to Kansei. Two experiments were performed to examine the proposed framework. Five females participated in the experiments. First, we measured participants' brain activities using an evaluation system that simulated product presentation and examined whether the estimator built from these data could estimate the precise preference levels. The stimuli were dress designs with various colors and shapes. The subjects' brain activities and button responses were obtained in two sessions. The estimator that outputs a preference level on the basis of brain activity patterns was built from the result of the first session. By applying the estimator to the former and the later brain information and by comparing the results and button responses, it was demonstrated that multilevel preference could be estimated. In the second experiment, the dress selection simulation using an IGA was performed. The IGA used the evaluation values which were obtained from the estimator constructed from the first session. They increased by generation achieved statistical significance. I was determined that preference levels could be estimated adequately using brain activities when similar stimuli were presented, e.g. a recommendation system for dress selection. The results contribute to the application of brain information to a product recommendation system based on IGA.