As curation depends on the experience and preference of the curator, visitors’ movements and locations can vary according to the planning curator, which results in different reactions from visitors. Unfortunately, it is unfeasible for curators to assess their exhibitions before the opening of the exhibition to check visitors’ reactions. This study proposes an algorithm that automatically recommends the structure of the museum and the locations of the exhibitions to reflect visitors’ preferences and the curator’s intention on the uncontrollable architecture of the showing room and the fixed set of artworks. The proposed algorithm uses a reinforcement-learning-based scheme to solve complicated problems by determining the best sequence of simple actions, which are scored based on multiple rules. The exhibition curated using the proposed algorithm was demonstrated and published as a virtual museum using Unity and WebGL and showed good effectiveness.