BehaviorFormer identifies distinct player actions in Minecraft gameplay
BehaviorFormer: Discovering Behavioral Patterns via Clusterable Next-Token Prediction
Machine LearningArtificial IntelligenceHuman-Computer Interaction
Summary
Finding patterns in how people play open-ended games is hard because there are no clear goals or labels. The authors combined a language prediction AI with a clustering method to group player actions into distinct categories. Using Minecraft data from 26 players, they found 23 different behavior patterns that represent common activities like mining and farming. These patterns help turn complex player actions into easy-to-understand groups.
What this means in practice
- •For game designers: Analyze player behaviors to tailor game mechanics and improve user experience by understanding distinct activity patterns.
- •For player behavior analysts: Automatically identify and categorize player actions in open-ended games for large-scale behavioral studies without manual labeling.
Tested on one dataset.
Authors
Tobias Schneider, Danial Fahrner, Soren Antebi, Daniel Uedelhoven, Lennard Bodden, Sandra Halscheidt, Rafet Sifa
Abstract
Discovering behavioral patterns in open-ended games is challenging due to the lack of clear objectives or ground-truth annotations. In this work, we extend a next-token prediction Transformer with the deep clustering algorithm DCSS to learn meaningful representations with well-separated clusters. In a controlled Minecraft session with 26 players, the method discovers 23 distinct behavioral patterns that generalize across the studied players, further refining structures present in self-supervised representations into clearly separable clusters. We group the clusters into seven interpretable categories covering stationary, roaming, travel, mining, base work, logging, and farming behavior.