System optimization without numerical target

Describes the importance of the challenging task of system optimization based on human preferences or subjective evaluation, and shows how to approach the task. As an example of this type of approach, an interactive genetic algorithm (GA) is introduced, and its advantages and disadvantages are discussed. To solve the disadvantages, this paper proposes two methods to reduce the burden of interactive GA operators, and evaluates their performances. Subjective tests have shown that the proposed methods are significantly better for human operators than conventional interfaces from a statistical test point of view.