Abstract
In this study, a music recommendation system that uses Kansei agents and music fluctuation properties is proposed. The objective is to search for music that matches users' preferences. Previous systems with comparable objectives did not adequately account for personal preferences in the recommendations that were returned. Kansei agents are Kansei model and characterized by a three-layered neural network. The neural network inputs correspond to user stimulus, and the output corresponds to the emotional response of the user because of the stimulus. Music fluctuation properties are defined as inputs in the neural networks of the Kansei agents. The proposed system returns recommendation based on personal subjectivity. The simulation that was conducted as a part of the study validated the effectiveness of the proposed system (i.e., the Kansei agents learned the users' Kansei to an acceptable level of accuracy).
