A proposal of interactive Tabu Search with paired comparison and differential vector for creating fragrance

Interactive evolutionary computation (IEC) is known as a method to optimize media contents suited to user's subjective feeling and preference. Previous IECs employed various evolutionary algorithms, and some of them applied Tabu Search (TS) algorithm: This method was named Interactive Tabu Search (ITS). In the ITS, users have to select the best individual from current population. ITS was often used for the area of computer graphics, and some previous studies applied ITS for creating fragrance. In these studies, blended fragrances composed of several aroma sources are corresponded to individuals in ITS. Adjusting intensity of each aroma source was target of optimization. Purpose of this study is to propose ITS that combines paired comparison in user's evaluation task and differential vector between the best individuals of different generations. The best individuals here mean that the best individuals in the current generation and in the previous generation. By combining these factors, we expect both of easy user's selection of the best individual and efficient search in searching fragrance. To investigate a fundamental efficiency of the proposed ITS, a smelling experiment was conducted. Target fragrance was a fragrance suited to a deodorant which has originally no fragrance.