Leveraging Convolutional Neural Networks for Evaluating Horror Games: A Novel Approach for Capturing Player Experience and Feedback

Abstract

Horror games stand out as one of the most viscerally immersive forms of content in the rapidly evolving space
of interactive entertainment. Understanding and improving the horror game player experience therefore requires
a methodical approach and employs metrics that can characterise the harnessing of various emotions and
perturbations elicited by these titles. Traditional assessment approaches often rely on qualitative measures that
are often highly subjective and hard to quantify. In this paper, we present a new method to assess horror games
using real-time player reaction analysis with convolutional neural networks (CNNs). We introduce a model that
utilises Convolutional Neural Networks (CNNs) to analyse in-game behaviour, physiological data and facial
expressions to quantify experience with minimal bias. We train our model on biometric data derived from
instances of gameplay sessions and from offline video-clips of people’s faces as they play games. By correlating
these metrics with specific in-game events captured by game engines, our technology can detect trends and
predict levels of fright, worrying or engagement. We found that CNNs perform with high accuracy at
recognising and categorising emotions, offering an instrument to improve the work of horror game developers.
Moreover, our approach can be scaled to improve the player experience for other gaming genres with similar
data-driven recommendations. This research will ultimately enable better, more sophisticated and more flexible
game production methods for the creation of games that will draw us closer to fictional experiences and are
more remarkable than many of the games that are currently available.