Computational player modelling

Description

Build computational models of players' cognitive, affective and behavioural states from gameplay metrics, body signals and context, using theory-driven, data-driven or hybrid approaches.

Player modelling is the computational modelling of a player's cognitive, behavioural and affective states from interaction data, distinct from static profiling. Model-based (top-down) approaches start from a theory such as arousal and valence, Russell's circumplex, OCC, flow or Koster's fun, and test fit to observations. Model-free (bottom-up) approaches learn the mapping from data to annotated states by classification, regression or preference learning, or cluster when labels are absent; most practice is hybrid. Inputs include gameplay metrics, physiological signals (ECG, GSR, EEG, EMG, pupillometry, gaze, posture, face), game context and profile. Outputs come from self-report, third-person annotation or unsupervised learning. Context fusion is essential: a GSR rise after a death and after level completion mean different things. Applications are adaptive difficulty, personalised content, believable agents, playtest analysis and monetisation.

Attributes

Loss type
Siloed
Data
Mixed
Effort
Months

Steps

  1. Decide which player states to model: cognitive, affective or behavioural.
  2. Choose a model-based, model-free or hybrid approach.
  3. Collect inputs: gameplay metrics, objective bodily signals, game context and any static profile.
  4. Obtain outputs through self-report, third-person annotation (ratings, classes or preferences) or, without labels, unsupervised clustering.
  5. Fuse game context with signals so that states are interpreted correctly.
  6. Train or fit the model (classification, regression, preference learning) and evaluate it on held-out players.
  7. Use the model to adapt or analyse the game, monitoring players' responses to adaptations and correcting misjudged states.

Sources

  • Player Modeling (Yannakakis et al., 2013)

Also known as

  • Computational player modelling (model-based, model-free and hybrid)
  • Player modeling
  • Player experience modelling
  • Experience-driven player modelling
  • Top-down vs bottom-up player modelling

Categories

  • Analytics & Behaviour
  • Computational & Machine Learning