2022
Roohi, Shaghayegh
Advances in AI-assisted game testing Väitöskirja
Tietotekniikka, Aalto-yliopisto, 2022, ISBN: 978-952-64-0825-5.
Abstract | Links | BibTeX | Avainsanat: AI agents, emotion recognition, englanninkieliset väitöskirjat, game testing, player experience, reinforcement learning
@phdthesis{Roohi2022,
title = {Advances in AI-assisted game testing},
author = {Shaghayegh Roohi},
url = {http://urn.fi/URN:ISBN:978-952-64-0825-5},
isbn = {978-952-64-0825-5},
year = {2022},
date = {2022-01-01},
school = {Tietotekniikka, Aalto-yliopisto},
abstract = {Game testing is an essential part of game development, in which developers try to select a game design that delivers a desirable experience for the players and engages them. However, the interactive nature of games makes the player experience and behavior unpredictable. Therefore, game testing requires collecting a large amount of playtest data in iterative sessions, which makes game testing time and money consuming.
Game testing includes a wide range of aspects from finding bugs and balancing game parameters to modeling player behavior and experience. This dissertation mostly concentrates on the player experience aspect. It proposes methods for (partially) automating and facilitating the game testing process. The first part of the dissertation focuses on player emotion analysis and proposes tools and methods for automatically processing and summarizing human playtesters' data. The second part of the dissertation concentrates on simulation-based approaches for modeling player experience and behavior to reduce the need for human playtesters.
In the first publication, we use deep neural networks for analyzing player facial expression data and provide a visualization tool for inspecting affect changes at game events, which replicates earlier results of physiological emotion analysis. Next, we extend this work by introducing a new dataset of game streamers' emotions in different granularities and considering other input signals like audio and speech for automatic emotion recognition. In the second part of the dissertation, simulation-based methods and reinforcement learning agents are used to predict game difficulty and engagement and capture the relation between these two game metrics.
In summary, this dissertation proposes and evaluates methods that advance automatic game testing by proposing approaches for automatic analysis of player emotions, which can be used for selecting specific segments of playtest videos for further inspection. In addition, we have provided accurate models of player experience and behavior using simulation-based methods that can be used to detect problematic game levels before releasing them to the actual players.},
keywords = {AI agents, emotion recognition, englanninkieliset väitöskirjat, game testing, player experience, reinforcement learning},
pubstate = {published},
tppubtype = {phdthesis}
}
Game testing is an essential part of game development, in which developers try to select a game design that delivers a desirable experience for the players and engages them. However, the interactive nature of games makes the player experience and behavior unpredictable. Therefore, game testing requires collecting a large amount of playtest data in iterative sessions, which makes game testing time and money consuming.
Game testing includes a wide range of aspects from finding bugs and balancing game parameters to modeling player behavior and experience. This dissertation mostly concentrates on the player experience aspect. It proposes methods for (partially) automating and facilitating the game testing process. The first part of the dissertation focuses on player emotion analysis and proposes tools and methods for automatically processing and summarizing human playtesters' data. The second part of the dissertation concentrates on simulation-based approaches for modeling player experience and behavior to reduce the need for human playtesters.
In the first publication, we use deep neural networks for analyzing player facial expression data and provide a visualization tool for inspecting affect changes at game events, which replicates earlier results of physiological emotion analysis. Next, we extend this work by introducing a new dataset of game streamers' emotions in different granularities and considering other input signals like audio and speech for automatic emotion recognition. In the second part of the dissertation, simulation-based methods and reinforcement learning agents are used to predict game difficulty and engagement and capture the relation between these two game metrics.
In summary, this dissertation proposes and evaluates methods that advance automatic game testing by proposing approaches for automatic analysis of player emotions, which can be used for selecting specific segments of playtest videos for further inspection. In addition, we have provided accurate models of player experience and behavior using simulation-based methods that can be used to detect problematic game levels before releasing them to the actual players.
Game testing includes a wide range of aspects from finding bugs and balancing game parameters to modeling player behavior and experience. This dissertation mostly concentrates on the player experience aspect. It proposes methods for (partially) automating and facilitating the game testing process. The first part of the dissertation focuses on player emotion analysis and proposes tools and methods for automatically processing and summarizing human playtesters' data. The second part of the dissertation concentrates on simulation-based approaches for modeling player experience and behavior to reduce the need for human playtesters.
In the first publication, we use deep neural networks for analyzing player facial expression data and provide a visualization tool for inspecting affect changes at game events, which replicates earlier results of physiological emotion analysis. Next, we extend this work by introducing a new dataset of game streamers' emotions in different granularities and considering other input signals like audio and speech for automatic emotion recognition. In the second part of the dissertation, simulation-based methods and reinforcement learning agents are used to predict game difficulty and engagement and capture the relation between these two game metrics.
In summary, this dissertation proposes and evaluates methods that advance automatic game testing by proposing approaches for automatic analysis of player emotions, which can be used for selecting specific segments of playtest videos for further inspection. In addition, we have provided accurate models of player experience and behavior using simulation-based methods that can be used to detect problematic game levels before releasing them to the actual players.