Modeling Individual and Team Behavior through Spatio-temporal Analysis
This work addresses the need for interpretable behavior modeling in multiplayer games, which has applications in areas like learner modeling and strategy analysis, but it is incremental as it builds on existing techniques like Dynamic Time Warping and clustering.
The paper tackles the problem of modeling player behaviors in games by introducing Interactive Behavior Analytics (IBA), a methodology with visualization systems, labeling, and algorithms, and demonstrates its effectiveness in developing human-interpretable models for team and individual behavior using data from BoomTown and DotA 2.
Modeling players' behaviors in games has gained increased momentum in the past few years. This area of research has wide applications, including modeling learners and understanding player strategies, to mention a few. In this paper, we present a new methodology, called Interactive Behavior Analytics (IBA), comprised of two visualization systems, a labeling mechanism, and abstraction algorithms that use Dynamic Time Warping and clustering algorithms. The methodology is packaged in a seamless interface to facilitate knowledge discovery from game data. We demonstrate the use of this methodology with data from two multiplayer team-based games: BoomTown, a game developed by Gallup, and DotA 2. The results of this work show the effectiveness of this method in modeling, and developing human-interpretable models of team and individual behavior.