A flowchart. A game screenshot is on top, and there are three columns underneath. The left column shows a player loop - seeing the game, experiencing emotion, making decisions, playing the game. The second shows an audience of play - seeing the gameplay and perceiving an emotion. The third demonstrates the PreGLAM model's position - watching gameplay variables, and modeling perceived emotion

PreGLAM: The Predictive, Gameplay-Based, Layered Affect Model

/ Uncategorized / By

Modeling a game spectator’s perceived emotion

A flowchart. A game screenshot is on top, and there are three columns underneath. The left column shows a player loop - seeing the game, experiencing emotion, making decisions, playing the game. The second shows an audience of play - seeing the gameplay and perceiving an emotion. The third demonstrates the PreGLAM model's position - watching gameplay variables, and modeling perceived emotion
A flowchart demonstrating how PreGLAM attempts to replicate the audience’s perceived emotion, rather than the player’s perceived or felt emotions

Abstract: We present the Predictive Gameplay-based Layered Aect Model (PreGLAM), an affective game spectator model. PreGLAM extends affective NPC emotion models to model a passive, biased spectator of gameplay. We implement PreGLAM into a custom game Galactic Defense, which we also describe. We empirically evaluate PreGLAM’s application in Galactic Defense, where we compare PreGLAM annotations with participant-provided ground-truth annotations. PreGLAM’s significantly outperforms a random walk time series in how accurately it matches ground-truth annotations.

The Predictive Gameplay-based Layered Affect Model, or PreGLAM, models a spectator’s perception of game emotion in real time, using a Valence-Arousal-Tension (VAT) model of emotion. PreGLAM bases its model of gameplay off of the interactions of the designed game mechanics during gameplay.

We use PreGLAMs model of a game spectator’s perceived emotion to control an adaptive musical score, extending Winifred Philips metaphor of game music acting “as an audience”. PreGLAM is based on an appraisal model of emotion, where emotions are based on how emotionally evocative events will affect the subject. We implement PreGLAM to essentially root for the player to win, and model the perceived emotions of game events (such as landing an attack, getting hit, healing the player) based on how the mechanics will move the player towards or away from victory.

This design allows PreGLAM to be broadly and flexibly integrated. Game designers regularly consider the ways that game mechanics will combine into player experiences, across broad and complex ranges of game genres, themes, and mechanics. PreGLAM formally represents the designed experience as a real-time model, based on the specific combinations of the mechanics during play. While we integrate PreGLAM into a combat-focused action game, the framework can be applied to a board range of game mechanics.

When evaluated, PreGLAM generally succeeds at modeling player emotion in real-time. While absolute measures of emotion are difficult to represent, PreGLAM’s output more closely matches ground-truth participant annotations of emotion, when compared with a random walk time series (no model, but motion and change)