Metaheuristic approach on temporal pattern matching for box office prediction

Abstract

It is important to understand customers (e.g., preferences) for box office prediction. Recently, many studies have focused on analyzing SNS data to predict box office. However, the studies have problems as follows: 1) decreasing the performance of prediction by the characteristics of SNS, and 2) increasing computational cost of prediction which is generated by processing SNS data. To solve the problems, we propose a metaheuristic method according to time-series data. The method generates three models of box office for one week after opening of a film, and recommends the optimal release timing based on the models for specific movie. For experiment, we collected box office data for 183 movies from Korean Film Council (KOFIC). As a result, proposed method can reflect effect of SNS and recommend optimal release date.