大型社群網絡之模式挖掘：從網絡架構至使用者行為

Pattern Mining in Large Social Networks: from Network Structure to User Behavior

羅經凱(Jing-Kai Lou)

指導教授 ： 雷欽隆

共同指導教授 ： 陳昇瑋(Sheng-Wei Chen)

國立臺灣大學/電機資訊學院/電機工程學研究所/博士(2014年)

https://doi.org/10.6342/NTU.2014.00763

摘要

隨著近年網絡的普及發達，線上社群媒體服務在這短短數年內變得欣欣向榮。透過社群媒體服務，人們在網絡上傳達思想或發展社交圈，變得前所未有的簡單。在社群網絡服務的推波助瀾下，如Facebook與Twitter，普羅大眾得以在網 絡世界中，展現自我，強化彼此關係，甚至形成線上社群。社群網絡的崛起， 打破了以往人們的交流模式，不再受到地理因素所限，使得這世界更加的緊密 相連。人們在網絡上所留下的足跡，成了最佳探討人類社交行為的豐富寶藏。 ! 長年累積的社交行為記錄，不僅為數眾多且種類極為多樣。學者們透過網絡架 構呈現巨大的社交行為資料，以人為點，人們之間的互動或者關係為連結，洞 察分析其中的脈絡。這一類的分析便稱作Social Network Analysis。透過這類分 析方式所探究的主題相當廣泛。大致上可以分為二類:其中一支主要的研究推 測社群網絡
(Social Network)的網絡結構與組織階層，某種程度上反應了人類 的社交行為，主要依循社群網絡結構探討人類社交行為。另外一類，則依循人們的屬性進行探討。我們則兩者併行，分析探討社群網絡議題。

我主要分析的對象為線上遊戲，其主要因素有二:一是線上遊戲自網絡發達以來，即視做殺手級應用，是人們上網的主要因素之一。二則是社群網絡提供了如現實世界般的投影，任由使用者沉浸其中遨遊，與他人進行各類互動。這提供讓我們同時觀察玩家個別或者社交行為的環境。

這份研究主要分為兩個部分探討。在第一個部分，我們觀察社群網絡架構形成的過程與探討形成的因素，藉此發現所得，套用於實際之應用。第二部分則改由觀察線上玩家的屬性與互動的關聯。我們充分利用所得之發現，用以預測一款遊戲的生命週期，及驗證社會心理學理論。

這篇論文的主要貢獻簡要列舉如下:我們對多款長壽的線上遊戲的玩家，進
行長期且通盤的行為與社交觀察。尤其所得的發現，我們開發出以下實際應
用。 1)依據社群網絡上的動態，我們提出一個方法快速估計個體在網絡中
的重要性。 2)我們設計一高維度信息擴散模型，用以模擬個體如何在群體
中發展影 響力。同時證明其收斂和其他幾個重要的特質。 3)我們蒐集玩家
遊戲時的情 緒生理反應，開發模型預測該遊戲上癮程度，間接預測遊戲生
命週期。 4)我們探討“性別交換”，這現象是指玩家選擇性別不同於玩家本
身性別的遊戲角色。我們報告的一款知名遊戲童話(Fairyland Online)在線
期間社會交往作為 玩家在遊戲中觀察到的行為模式，並討論在虛擬社交場
合的性別角色和自我形象的影響。

關鍵字

使用者行為 ； 社群網絡分析 ； 網絡測量 ； MMORPG

並列摘要

Social media has prospered in the last decade, which grants people a hitherto wide range to deliver their views, and to develop their social circles through the Internet. The worldwide popular social media services such as Facebook, facilitate people to manifest themselves, to engage the connections, and to join communities. Rise of social media unleashes people from the geographic limitation, and makes the world more connected. The users globally form an online society via the social media services. Thank to the trackable online activities, the interactions on social media services can be completely saved, which leaves us a hitherto abundant resource to study social behaviors of humans.

The tremendous and various data in social media greatly draws attention to the analysis of the social network. A social network refers to a network structure consisting of the actors on social media. Research interests on social network analysis are widely diverse. Generally speaking, one primary study is the social network structuralism that postulates the network structure implicitly reflects the human behavior. The interests would be narrow down to the relevance between the structural patterns and social issues. On the other hand, another research line particularly devotes to the content or the attributes of the social actors. We endeavoured to explore the issues on social media in both aspects.

We aim at the social online gaming as our main target to study for two reasons. First, online gaming is considered as a killer application since the emergence of World Wide Web, and has been part of the primary reasons people use the Internet. Second, the realistic nature of the interactions in online gaming opens up the possibility of viewing the world inside an MMORPG as a laboratory for observing individual and social behaviors of humans in as fine details as the game design would allow.

The work is divided into two parts. In the first part, we investigate the cause of the unique social network structure. The findings from the observations inspire us a number of practical applications. In the second part, we shift our focus to the attributes of the human behavior in online gaming which functions as a virtual society. We exploit the observations to predict the duration of an online game, and also to confirm the sociological and psychological theories.

The contributions of the dissertation are briefly listed as follows. We perform a comprehensive observation of the individual activities and social interactions in a number of long-lived online games. With the knowledge gained from the investigation, we develop the following applications for academic or industrial purposes. 1) We propose an approach to fast estimate individual centrality according to the dynamics of a social network. 2) We design a preference diffusion model with theoretical guarantees to manage high dimensional information. We assess the quality of this model by proving its convergence and several other important properties. 3) We propose a quantitative way for the addictiveness index of a game according to the gamers' online sessions. Then, we develop a forecasting model for the addictiveness index of a game not released in the market yet. 4) we investigate the phenomenon of ``gender swapping,' which refers to players choosing avatars of genders opposite to their natural ones. We report the behavioral patterns observed in players of Fairyland Online during social interactions when playing as in-game avatars of their own real gender or genderswapped, and discuss the effect of gender role and self-image in virtual social situations.

並列關鍵字

user behaviour ； social network analysis ； complex network ； network measure ； MMORPG

參考文獻

[1] Linton C. Freeman. A set of measures of centrality based on betweenness. Sociometry, 40(1):35–41, March 1977.

[2] Ulrik Brandes. A faster algorithm for betweenness centrality. Journal of Mathematical Sociology, 25:163–177, 2001.

[4] Duncan J. Watts and Steven H. Strogatz. Collective dynamics of ’small-world’ net- works. Nature, 393(6684):440–442, June 1998.

[5] MEJ Newman. A measure of betweenness centrality based on random walks. Social networks, 27(1):39–54, 2005.

[6] Alain. Barrat, Marc Barthelemy, Romualdo. Pastor-Satorras, and Alessandro. Vespig- nani. The architecture of complex weighted networks. Proceedings of the National Academy of Sciences, 101(11):3747, 2004.