Discussion of the crossover method of interactive Genetic Algorithm for extracting multiple peaks on Kansei landscape

Interactive Genetic Algorithms (iGAs) are optimization techniques used to estimate customers´ Kansei (Japanese term for computing that relates to human characteristics such as sensibility, perception, affection or subjectivity) because human subjective evaluations are replaced with the objective function of Genetic Algorithms (GAs). Applying iGAs to recommend a product to a customer is examined in our study. One of the requirements is to estimate multiple preferences of a user and reflect preferences in the recommended products shown to him or her. When users select their preferred products within a specific category, they might like various kinds of products. In our study, these preferences are defined as multimodal preferences. When searching products a user would want, the recommendation method displays the more favored products by considering this multimodal preference. Therefore, in this study, we discuss using an iGA to generate offspring by estimating and searching multiple peaks. Our proposed method estimates multiple peaks by clustering the parents that the customer has evaluated more favorably and generates the appropriate offspring by constructing the probabilistic model based on the distribution of parents within a cluster. We performed two experiments. In the first experiment, we confirmed that the participants of the experiment had multimodal preferences. In the second experiment, the participants operated one of two systems which implemented either the proposed method or conventional method. The comparison of results showed that the system that implemented the proposed method searched the participants´ multimodal preferences more diversely than the system that implemented the conventional method.