Games and play have become a primary cultural form in modern human society. Game culture is data-intensive, and its landscape has become more complex and enriched by social media.
Players leave their ‘footprints', the data players have left in social media, online for example, by leaving game reviews and comments, showcasing accomplishments, building communities, and engaging in discussions with other players to express their opinions.
Those player-generated contents are valuable resources and can contribute to a richer understanding of games and play.
“By understanding players, we can for example probe player experiences by analyzing their written reviews. We can also understand player without conducting another interview or survey”, MSc Chien Lu says.
In his dissertation, Lu investigates the potential of representation learning, and how to effectively harness such data, which is often intricate and with complex structures.
“This approach leverages machine learning algorithms to disentangle data dependencies, model temporal dynamics, learn diverse representations, and tackle heterogeneous data. By integrating empirical analysis and methodology development, the work underscores the positive impact of representation learning in game studies”, Lu describes.
Illustrating the capacity of representation learning
To uncover insights in game studies, the study illustrates the capacity of representation learning through exploring player typologies and modeling of the evolution of player perceptions. It facilitates a more flexible framework for player typologies and enhances the understanding of the evolution of player experiences over time.
Inspired by empirical analysis, the dissertation further develops representation learning methodologies.
“The developed factor model for text enables modeling the temporal dynamics of textual data from multiple sources. The integration of a factor model into a topic model allows a comprehensive examination of the influence of player characteristics on their game reviews”, Lu explains.
Furthermore, the developed embedding models offer new opportunities when analyzing online player communities, including Twitch streamers and Reddit online communities. The Bayesian non-parametric embedding model facilitates learning more diverse representations over player networks. The Gaussian copula-based model enables learning representations in heterogeneous data settings.
"Through pursuing two research directions, empirical data analysis and methodology development, my research can facilitate interdisciplinary collaboration and further benefit the research communities”, Lu says.
Public Defence on Wednesday 6 September
The doctoral dissertation of MSc Chien Lu in the field of Machine Learning titled Representation learning techniques are proved useful for game data analysis will be publicly examined at the Faculty of Information Technology and Communication Sciences at 13 o'clock on Wednesday 6 September, 2023. The venue is at the City centre campus, Linna Building lecture hall K103, Kalevantie 5. The opponent will be Professor Arto Klami from the University of Helsinki. The custos will be Professor Jaakko Peltonen from Tampere University.
The dissertation is available online.
The public defence can be followed via remote connection.
Photo: Sari Laapotti