Summary
Player modelling is the practice of building computational models of what players do, prefer, know, or feel. These models can be used to personalise content, estimate skill, predict churn, adapt difficulty, or study player experience. In AI-and-games research, player modelling is treated as a first-class area of game AI alongside game playing and procedural content generation. In practice, it usually begins with telemetry rather than with a complete psychological theory. (Yannakakis and Togelius, Artificial Intelligence and Games, see source-ai-and-games)
Key ideas
- Top-down models: start from a theory or taxonomy, then fit data into it.
- Bottom-up models: infer clusters or patterns directly from telemetry.
- Inputs: gameplay traces, objective performance, context, profile data.
- Outputs: predicted behaviour, predicted experience, or no direct output beyond analysis.
In practice
Common game uses:
- Estimate whether a player is novice, competent, or expert.
- Predict whether the player is frustrated, engaged, or likely to leave.
- Adapt encounter difficulty, tutorial pacing, or content recommendations.
In a Unity analytics workflow, a player model usually starts as feature engineering rather than as exotic AI:
- completion time
- deaths or retries
- weapon or build preferences
- exploration percentage
- input rhythm or accuracy
Evidence
- Artificial Intelligence and Games distinguishes modelling behaviour from modelling experience, which is important because a player can perform well while still having a poor experience.
- The same source also distinguishes model-based and model-free approaches, which maps well onto the difference between theory-led design and pure telemetry mining.
Implications
- Player modelling connects the programming/AI layer to game-analytics, flow, self-determination-theory, and adaptive design decisions.
- It is easy to overclaim. A telemetry model usually predicts proxies, not the full inner life of the player.
Open questions
- Which player experience variables are robust enough to model from classroom-scale telemetry?
- How should the vault connect player modelling to ethical concerns around manipulation and dark patterns?
Related
game-analytics · telemetry · machine-learning-games · procedural-generation · ai-experience-management · overview-cre341-agent-ai-route · overview-artificial-intelligence-in-games · source-ai-and-games