Personalized Interactive Narratives via Sequential Recommendation of Plot Points

Abstract

In story-based games or other interactive systems, a drama manager (DM) is an omniscient agent that acts to bring about a particular sequence of plot points for the player to experience. Traditionally, the DM's narrative evaluation criteria are solely derived from a human designer. We present a DM that learns a model of the player's storytelling preferences and automatically recommends a narrative experience that is predicted to optimize the player's experience while conforming to the human designer's storytelling intentions. Our DM is also capable of manipulating the space of narrative trajectories such that the player is more likely to make choices that result in the recommended experience. Our DM uses a novel algorithm, called prefix-based collaborative filtering (PBCF), that solves the sequential recommendation problem to find a sequence of plot points that maximizes the player's rating of his or her experience. We evaluate our DM in an interactive storytelling environment based on choose-your-own-adventure novels. Our experiments show that our algorithms can improve the player's experience over the designer's storytelling intentions alone and can deliver more personalized experiences than other interactive narrative systems while preserving players' agency.