Predicting Players' Emotions from Game Telemetry

MSc Thesis Defence: Ingibjörg Ósk Jónsdóttir

  • 15.8.2017, 12:00 - 13:00

Time: August 15th, 12am
Room: M209
Student: Ingibjörg Ósk Jónsdóttir
Thesis Title: Predicting Players' Emotions from Game Telemetry
Supervisor: David Thue 

Abstract: In video games, interactive storytelling systems often use a player model to tailor the storyline to a player's preferences, personality, or skills. The player's in-game actions are often used as an input for the model, but their emotions can also offer useful knowledge. People often find it hard to describe their emotions and therefore, we aim to measure them through their physiological response. Since players are usually not equipped with such devices in their natural gaming environment, we seek to render the devices unnecessary by developing a method to predict a player's emotions from their in-game actions. Our method involves a user study where the player's actions are tracked and their physiological response is recorded. We then compute three emotion features (arousal, valence, and dominance) and train several machine learning algorithms to predict those features from the player's in-game actions. Our results show that our method can predict a player's emotion features from their in-game actions more accurately than the results of a uniform random predictor.