Investigation of the efficiency of continuous evaluation‐based interactive evolutionary computation for composing melody

Abstract

Interactive Evolutionary Computation (IEC) is known as a method reflecting user's feelings on media contents. However, user's fatigue caused by repetitive evaluation tasks remains as a problem. The authors proposed continuous evaluation‐based IEC (CEIEC) that divides user's evaluation tasks into multiple days, while users finish the task in 1 day in conventional IECs. The efficiency of CEIEC was shown by the 2‐day experiment: the highest fitness was obtained in the final generation of the second day. The purpose of this study is to show further efficiencies of CEIEC by comparing experiment. The target of search was designing a 'bright' sign sound. Forty‐two subjects were divided into two groups: half of them performed the 2‐day task of CEIEC, and the other half performed the 1‐day task of IEC. After that, they evaluated representative three sounds. A gradual decrease in update of the solution candidates and a shrink of the search space were observed in both conditions. Furthermore, the same trend of increases in fitness was observed. These results showed the efficiencies of CEIEC in terms of having the same searching ability as the ordinal IEC. Additionally, elapsed time for completing the task was significantly shorter in CEIEC. © 2019 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.