摘要
隨著網路遊戲的日漸熱門，越來越多玩家透過網路遊戲相關的社群媒體，討論遊戲內容或者分享自己的遊戲心得。這些社群媒體上的內容可以提供有用的資訊給遊戲公司以及玩家做參考；然而，這些內容的資料量非常龐大且包含大量無用的資訊。因此，本論文提出一個研究架構，將遊戲論壇及Facebook專業的資料分群成多個主題，並且針對每一個主題，分析其隨時間的演化、情緒分布及跨社群的比較，提供洞見給遊戲公司及玩家。此架構包含四個部分，首先，針對每一討論串取出遊戲相關的字詞形成虛擬文件；接著，修改隱含狄利克雷分布分群方法，加強字詞共同出現的重要性將虛擬文件分群成多個主題，並且將每一主題再依其中字詞的時間戳記分群成數個子主題；然後，計算每一主題及討論串之情緒分布並且找出異常值；最後，分析及比較跨社群的類似主題，提供不同面向的資訊予遊戲公司。實驗結果顯示，我們的研究架構能將相似且相關聯的字詞分群在同一主題中，且主題演化、情緒及跨平台分析之結果，能幫助遊戲公司了解玩家需求及協助玩家了解遊戲玩法的攻略及指引。
並列摘要
As online games have become more and more popular recently, more and more players discuss and share their opinions on online game social media. The contents of media are informative and helpful to game companies and players, but they may be overwhelming and contain plenty of noises. Moreover, every social media may provide different aspects of information. Therefore, in this thesis, we propose a framework to cluster the contents in the forum and the Facebook page of an online game into several topics, analyze the evolution of each topic and sentiment distribution, and compare the topics clustered on both media to provide helpful insights for game companies and players. The proposed framework contains four phases. First, we extract game terms from threads to form a virtual document, where a thread contains a question and a sequence of comments about the question. Second, we modify Latent Dirichlet Allocation (LDA) to cluster the virtual documents into latent topics by increasing the importance of co-occurred game terms. For each topic, we further divide it into several sub-topics by considering the timestamp of each game term in the topic. Third, we compute the sentiment distribution of each topic and thread, and find the outliers. Finally, we analyze similar topics found from both social media, which can provide different insights for game companies. The experiment result shows that the proposed framework can cluster similar and relevant game terms together. The topic evolution, sentiment analysis and intermedia analysis can help game companies understand the demand of players, and help players to know the guidelines and tips of playing the game.