User evaluation prediction models based on conjoint analysis and neural networks for interactive evolutionary computation

The authors develop the user evaluation prediction models based on conjoint analysis and neural networks for interactive evolutionary computation (IEC) implemented by interactive genetic algorithm and interactive differential evolution. In addition, the facial expression generation system described in this paper simulates user evaluation based on personalized models and generates images of happy faces and sad faces automatically as an example. IEC that can optimize its targets according to the user’s preference and sensibility is attracting attention as an interactive personalization method. However, IEC has the problem of user evaluation fatigue because it requires a lot of user evaluations to search the optimum solution. Therefore, interactive systems employing IEC are used with a user evaluation prediction model so that they can reduce a user’s load. The novelties of this study are combination of conjoint analysis and large scale neural networks integrated with user evaluation prediction models. Finally, the authors verify usability of the proposed models by performing user evaluation experiments. As a result, the proposed models indicate better prediction accuracy of user evaluation than a previous research using a simple neural network. Also, the personalized models can simulate user evaluation successfully.