2023
Lu, Chien
2023, ISBN: 978-952-03-3031-6.
Abstract | Links | Tags: Game studies, Learning techniques, Shallow representation learning
@phdthesis{Lu2023d,
title = {Shallow Representations, Profound Discoveries: A Methodological Study of Game Culture in Social Media},
author = {Chien Lu},
url = {https://trepo.tuni.fi/handle/10024/150586},
isbn = {978-952-03-3031-6},
year = {2023},
date = {2023-09-06},
urldate = {2023-09-06},
publisher = {University of Tampere},
abstract = {This thesis explores the potential of representation learning techniques in game studies, highlighting their effectiveness and addressing challenges in data analysis. The primary focus of this thesis is shallow representation learning, which utilizes simpler model architectures but is able to yield effective modeling results. This thesis investigates the following research objectives: disentangling the dependencies of data, modeling temporal dynamics, learning multiple representations, and learning from heterogeneous data. The contributions of this thesis are made from two perspectives: empirical analysis and methodology development, to address these objectives. Chapters 1 and 2 provide a thorough introduction, motivation, and necessary background information for the thesis, framing the research and setting the stage for subsequent publications. Chapters 3 to 5 summarize the contribution of the 6 publications, each of which contributes to demonstrating the effectiveness of representation learning techniques in addressing various analytical challenges.
In Chapter 1 and 2, the research objects and questions are also motivated and described. In particular, Introduction to the primary application field game studies is provided and the connections of data analysis and game culture is highlighted. Basic notion of representation learning, and canonical techniques such as probabilistic principal component analysis, topic modeling, and embedding models are described. Analytical challenges and data types are also described to motivate the research of this thesis.
Chapter 3 presents two empirical analyses conducted in Publication I and II that present empirical data analysis on player typologies and temporal dynamics of player perceptions. The first empirical analysis takes the advantage of a factor model to offer a flexible player typology analysis. Results and analytical framework are particularly useful for personalized gamification. The Second empirical analysis uses topic modeling to analyze the temporal dynamic of player perceptions of the game No Man’s Sky in relation to game changes. The results reflect a variety of player perceptions including general gaming activities, game mechanic. Moreover, a set of underlying topics that are directly related to game updates and changes are extracted and the temporal dynamics of them have reflected that players responds differently to different updates and changes.
Chapter 4 presents two method developments that are related to factor models. The first method, DNBGFA, developed in Publication III, is a matrix factorization model for modeling the temporal dynamics of non-negative matrices from multiple sources. The second mothod, CFTM, developed in Publication IV introduces a factor model to a topic model to handle sophisticated document-level covariates. The develeopd methods in Chapter 4 are also demonstrated for analyzing text data.
Chapter 5 summarizes Publication V and Publication VI that develop embedding models. Publication V introduces Bayesian non-parametric to a graph embedding model to learn multiple representations for nodes. Publication VI utilizes a Gaussian copula model to deal with heterogeneous data in representation learning. The develeopd methods in Chapter 5 are also demonstrated for data analysis tasks in the context of online communities.
Lastly, Chapter 6 renders discussions and conclusions. Contributions of this thesis are highlighted, limitations, ongoing challenges, and potential future research directions are discussed.},
keywords = {Game studies, Learning techniques, Shallow representation learning},
pubstate = {published},
tppubtype = {phdthesis}
}
This thesis explores the potential of representation learning techniques in game studies, highlighting their effectiveness and addressing challenges in data analysis. The primary focus of this thesis is shallow representation learning, which utilizes simpler model architectures but is able to yield effective modeling results. This thesis investigates the following research objectives: disentangling the dependencies of data, modeling temporal dynamics, learning multiple representations, and learning from heterogeneous data. The contributions of this thesis are made from two perspectives: empirical analysis and methodology development, to address these objectives. Chapters 1 and 2 provide a thorough introduction, motivation, and necessary background information for the thesis, framing the research and setting the stage for subsequent publications. Chapters 3 to 5 summarize the contribution of the 6 publications, each of which contributes to demonstrating the effectiveness of representation learning techniques in addressing various analytical challenges.
In Chapter 1 and 2, the research objects and questions are also motivated and described. In particular, Introduction to the primary application field game studies is provided and the connections of data analysis and game culture is highlighted. Basic notion of representation learning, and canonical techniques such as probabilistic principal component analysis, topic modeling, and embedding models are described. Analytical challenges and data types are also described to motivate the research of this thesis.
Chapter 3 presents two empirical analyses conducted in Publication I and II that present empirical data analysis on player typologies and temporal dynamics of player perceptions. The first empirical analysis takes the advantage of a factor model to offer a flexible player typology analysis. Results and analytical framework are particularly useful for personalized gamification. The Second empirical analysis uses topic modeling to analyze the temporal dynamic of player perceptions of the game No Man’s Sky in relation to game changes. The results reflect a variety of player perceptions including general gaming activities, game mechanic. Moreover, a set of underlying topics that are directly related to game updates and changes are extracted and the temporal dynamics of them have reflected that players responds differently to different updates and changes.
Chapter 4 presents two method developments that are related to factor models. The first method, DNBGFA, developed in Publication III, is a matrix factorization model for modeling the temporal dynamics of non-negative matrices from multiple sources. The second mothod, CFTM, developed in Publication IV introduces a factor model to a topic model to handle sophisticated document-level covariates. The develeopd methods in Chapter 4 are also demonstrated for analyzing text data.
Chapter 5 summarizes Publication V and Publication VI that develop embedding models. Publication V introduces Bayesian non-parametric to a graph embedding model to learn multiple representations for nodes. Publication VI utilizes a Gaussian copula model to deal with heterogeneous data in representation learning. The develeopd methods in Chapter 5 are also demonstrated for data analysis tasks in the context of online communities.
Lastly, Chapter 6 renders discussions and conclusions. Contributions of this thesis are highlighted, limitations, ongoing challenges, and potential future research directions are discussed.
In Chapter 1 and 2, the research objects and questions are also motivated and described. In particular, Introduction to the primary application field game studies is provided and the connections of data analysis and game culture is highlighted. Basic notion of representation learning, and canonical techniques such as probabilistic principal component analysis, topic modeling, and embedding models are described. Analytical challenges and data types are also described to motivate the research of this thesis.
Chapter 3 presents two empirical analyses conducted in Publication I and II that present empirical data analysis on player typologies and temporal dynamics of player perceptions. The first empirical analysis takes the advantage of a factor model to offer a flexible player typology analysis. Results and analytical framework are particularly useful for personalized gamification. The Second empirical analysis uses topic modeling to analyze the temporal dynamic of player perceptions of the game No Man’s Sky in relation to game changes. The results reflect a variety of player perceptions including general gaming activities, game mechanic. Moreover, a set of underlying topics that are directly related to game updates and changes are extracted and the temporal dynamics of them have reflected that players responds differently to different updates and changes.
Chapter 4 presents two method developments that are related to factor models. The first method, DNBGFA, developed in Publication III, is a matrix factorization model for modeling the temporal dynamics of non-negative matrices from multiple sources. The second mothod, CFTM, developed in Publication IV introduces a factor model to a topic model to handle sophisticated document-level covariates. The develeopd methods in Chapter 4 are also demonstrated for analyzing text data.
Chapter 5 summarizes Publication V and Publication VI that develop embedding models. Publication V introduces Bayesian non-parametric to a graph embedding model to learn multiple representations for nodes. Publication VI utilizes a Gaussian copula model to deal with heterogeneous data in representation learning. The develeopd methods in Chapter 5 are also demonstrated for data analysis tasks in the context of online communities.
Lastly, Chapter 6 renders discussions and conclusions. Contributions of this thesis are highlighted, limitations, ongoing challenges, and potential future research directions are discussed.