Facing Your Self: Dynamic Difficulty Adjustment in Hades via Deep Player Behavior Modeling

Abstract

This master's thesis explored the development and evaluation of enhancing Dynamic Difficulty Adjustment (DDA) with Deep Player Behavior Modeling (DPBM) in the context of video games. The DPBM approach of this thesis is designed to capture and replicate individual player behaviors, providing a personalized and engaging gaming experience. To test the effectiveness, I developed a mod named "Dark Zagreus" for the award-winning action roguelike game, Hades. In this modified version, the final boss is replaced with Dark Zagreus, an AI agent that leverages DPBM to learn from the player's last successful run and imitate player behavior. Within a study long lasted two weeks, 20 players ($n=20$) participated in an approximately one-hour session, where they encountered the boss in both scripted and DPBM modes and filled out a questionnaire. At the end of the study, a fluctuating result was observed and no significant differences regarding player experience metrics and imitation scores were found. In the open-ended responses to the questionnaire, a portion of participants noted that Dark Zagreus effectively mirrored their combat tactics, suggesting successful behavior replication. Regarding the performance, DPBM shows a comparable accuracy with similar previous works, which still brings this research valuable insight. Despite limitations such as the exclusion of certain game mechanics, this master's thesis demonstrated the potential of DPBM in advancing DDA research. Aside from this contribution, the source code of the mod is deployed publicly to support further research in this field, offering a valuable testbed for future studies aimed at refining DDA techniques and enhancing player engagement.