Modeling Director Agents' Decision-Making Strategies in Guided Discovery Learning Environments

摘要

Interactive narrative environments offer significant potential for creating engaging narrative experiences. Increasingly, applications in education, training, and entertainment are leveraging narrative to create rich interactive experiences in virtual storyworlds. A key challenge posed by these environments is devising accurate models of director agents' strategies that make director intervention and action decisions to craft customized story experiences for users. Director agents work behind the scenes to direct a cast of non-player characters and storyworld events for the unfolding narrative. Although a growing body of research has investigated techniques for modeling director agents in interactive narrative, prior work has focused on models learned from simulated data or pre-authored models. A promising approach is developing an empirically driven model of director agents' decision-making strategies.;In this work, we propose a dynamic Bayesian network framework for modeling director agent narrative decision-making. To create empirically informed models of director agent decision-making strategies, we conducted a Wizard-of-Oz study with an interactive narrative-centered learning environment. In the study, the wizard served as a "human director agent." Machine learning was used to automatically acquire the conditional probabilities for the dynamic Bayesian networks. The machine-learned models were then empirically evaluated to investigate their effectiveness and efficiency in real-time. Results of the study are encouraging and suggest that empirically driven models of director agent decision-making strategies can offer significant predictive power.