Predicting personality traits from how people play shooter games
- Article
- Psychology
Can the way you play a video game say something about who you are? A study by Muhannad Quwaider, Abdullah Alabed, and Rehab Duwairi, published in Simulation Modelling Practice and Theory in January 2023, set out to test this with machine learning.
Why a new game
Earlier work on games and personality leaned on the Myers-Briggs Type Indicator, which sorts people into one of two opposite types on each scale. The authors argue that this is too coarse. They used the Five-Factor Model instead, which scores five traits on a continuous range: openness, conscientiousness, extraversion, agreeableness, and neuroticism. The idea behind the study is simple. Behavior in a virtual world reflects behavior in the real one, so if you record enough of what a player does, a model should be able to estimate their traits.
Rather than reuse data from a commercial title, the team built their own first-person shooter, called The Protector, and released it on the Google Play Store. Players took a Five-Factor personality questionnaire, and the game logged what they did while playing. Because the game was public, the data came from thousands of real users in uncontrolled settings rather than from a small lab sample.
What the models did
The authors ran two kinds of models. The first was classification: given a player’s in-game behavior, predict their personality category. An artificial neural network did this best, reaching 98.6 percent accuracy on their data. The second was regression: given a player’s history, predict how they will act in future rounds of the game. Here support vector regression and k-nearest neighbors regression gave the lowest error.
Reading the result
A 98.6 percent figure is striking, and it is worth being careful with it. The result holds for one game, one set of logged features, and players who chose to install it. Whether the same signals transfer to other games or to people who do not play shooters is a separate question the study does not settle. Still, the work shows that gameplay logs carry enough signal to separate personality profiles, and that a purpose-built game is a workable way to gather that data at scale.