Using Supervised Machine Learning to Map In-Game Interactions to a Player Behaviour Model based on Game Dynamics
Authors
Faldt, Mathias Nedergaard Gydesen ; Johnsen, Christian Bank ; Edvars, No Alexander Gynther ; Sørensen, Kasper
Term
4. term
Education
Publication year
2021
Submitted on
2021-05-25
Pages
67
Abstract
The aim was to build a supervised machine learning model that helps video games adapt to each player’s preferences. Earlier work used unsupervised behavior models, which were hard for designers to apply because the relationships in the models were not transparent. We used the Core Game Dynamics (CGD) framework and a questionnaire to describe player behavior, and we built a CGD-based game designed to appeal to three player types: Assault, Manage, and Journey. We collected data from 102 participants and trained a supervised classifier offline (i.e., not during gameplay) to predict player type. The model’s overall accuracy was 48%. When we split the data into segments with different characteristics—specifically, by the average difference between a player’s scores for the three types—accuracy improved to 71%.
Formålet var at udvikle en superviseret maskinlæringsmodel, der kan få videospil til at tilpasse sig den enkelte spillers præferencer. Tidligere arbejde brugte usuperviserede modeller af spilleradfærd, men de var svære at bruge i designet, fordi sammenhængene i modellen ikke var tydelige. Vi anvendte Core Game Dynamics (CGD)-modellen og et spørgeskema til at beskrive spilleradfærd og byggede et spil baseret på CGD, der skulle appellere til tre spillertyper: Assault, Manage og Journey. Vi indsamlede data fra 102 deltagere og trænede derefter en superviseret klassifikationsmodel offline (altså ikke under selve spillet) til at forudsige spillertype. Modellen opnåede en generel nøjagtighed på 48%. Da vi delte data op i segmenter med forskellige kendetegn—specifikt efter den gennemsnitlige forskel mellem en spillers scores for de tre typer—steg nøjagtigheden til 71%.
[This apstract has been rewritten with the help of AI based on the project's original abstract]