Construction of the Aroma Therapist's Kansei Model Using a Neural Network

Abstract

Recently, worries about health and/or life stresses can cause serious issue for people. Various solutions have been proposed and tried to ease peoples' troubles. One way is through aromatherapy, which provides mental comfort for people under healthy conditions. Many therapists propose a suitable fragrance for a client by applying original therapy methods. However, they merely select an aroma using a convenience linkage such as “If A then B” and so on. Their methods often find it difficult to satisfy a client's needs. To resolve such issues, our objective is to model therapist's Kansei on a computer. Kansei looks like human experiences or sensibilities. In this study, we constructed a Kansei model for the aroma therapist using a neural network suitable for reproducing human sensibilities. The model is a three layered network consisting of twelve input and output nodes. The input is the point of a questionnaire classified into twelve alternatives. The output has twelve pairs of aromas selected beforehand by the counselor. As the result of their performance tests of the network after training, we experimentally confirmed that the output of the network almost matched the aromas chosen by the therapist.