A Supervised Learning Approach to Build a Recommendation System for User-Generated Content in a Casual Game
Paulo C. R. Souza Neto, Tulio Braga Moreira Pinto · 2017
There is a growing tendency in games that made use of usergenerated content, artifacts created for a game by the players instead of game developers. This type of content may allow games to have an extension of playable hours and then a greater longevity. However, allowing users to provide their own content may result in low-quality game experience to other players, caused by the difficulty to find a content that delight the user base that will consume the user-generated content. In this paper, we present an approach to creating pleasant content recommendations to the users. We propose the use of Supervised Learning to build a model capable of predicting positive ratings from a user over a specific content and use this information to build the recommendation system. The recommendation system is capable to handle each user individually and create their own content recommendation list. To validate and measure our strategy performance, we collected user ratings of stages created by others user in a mobile puzzle game called Mr. Square, during a period of 5 months. Our results show that we can improve the percentage of positively rated content, and then the user satisfaction, to 96.2%, while the current strategy implemented in Mr. Square holds a satisfaction of only 70.1%.