In our master thesis, we present a system for maintaining a user's engagement in computer games based on measured psychophysical indicators such as average heartbeat, galvanic skin response and electric activity in the brain. We wish to prove that hand adjustment of video game parameters can instead be performed by an automated system for maintaining player engagement. We split the implementation of such a system into two separate sections. The first section consists of a user-engagement model that predicts the current user engagement based on the player's gameplay characteristics and measured psychophysical indicators. We have built the engagement model using supervised machine learning techniques. The second section of the system is represented by an algorithm that, when a drop in user engagement is detected, adjusts game parameters in order to maintain user engagement. The optimal strategy for changing game parameters was learned through the use of reinforcement learning.
|
