Detecting Player Preference Shifts in an Experience Managed Environment

Players are often considered to be static in their preferred play styles, but this has been shown to be untrue. While this is not an issue in games that only attempt to craft a singular experience for the player, this can become a problem for games that use an Experience Manager (ExpM) to adapt the game to the player. As an ExpM observes the player’s actions within the game environment, it makes changes to the environment to bias it towards the player’s preferences, but if a player were to shift their preferences then any further ExpM observations are now based on this biased environment. These biased observations may not be able to show the player’s true preferences and cause the ExpM to form an incorrect player model. In this situation the ExpM should be able to detect these player preference shifts. In this paper we show that such a shift can be detected using only the observations of the player without further intervention. To evaluate these claims we use a text-based interactive fiction environment created to exhibit this bias caused by an ExpM. We compare our own method to detect such shifts to existing outlier detection methods and find that our method can detect shifts in player preferences.