社群網路服務的大量竄起，讓人們在網路上產生了許多交互影響的社群行為，其中以資訊分享及追隨行為最為廣泛，而資訊的型態多半是以文本的方式做傳遞，例如新聞、評論及活動消息等等，產生許多評估社群網路中個別使用者之影響力抑或最大影響範圍的研究，但其中卻缺乏對於訊息傳遞行為是如何形成、如何影響做更深入分析。 以往的影響力探討，多半是對於網路結構的剖析，對於訊息的性質或者是傳遞行為中使用者的屬性未有深入定義，即使有將使用者分群，也多是利用單面向角度思考，但社群網路應屬於一個異質性網路，使用者應在不同的影響關係呈現不同的身份，故本研究別於以往單分類使用者的方式，改以多隸屬興趣程度在不同階層判別使用者的角色，並對於訊息傳遞資料有進一步的探勘，試圖建置一個可從多構面分析文本訊息傳遞行為的工具，藉此分析者除了可以了解訊息傳遞的因由外，還能有效地應用至協同式推薦或口碑式行銷等。 本研究分為四個部分：第一部分為定義興趣社群階層，利用階層概念將興趣從細膩到廣泛的作定義。第二部分為根據使用者過去發表的文章及活動紀錄(如轉貼或者追隨等)，歸屬使用者較具有代表性的興趣主題群，第三部分為萃取有效的訊息傳遞以客觀地判斷使用者彼此的交互影響行為。第四部份為建置分析平台，針對社群行為之訊息傳播，提出多種分析功能並利用圖表方式讓分析者能快速掌握資訊結果。 With the rise of social network service, there are many social behaviors. The most popular behaviors are information sharing and following. Users share their ideas and interesting things with their friends on social network sites. Moreover, users select some people to follow and obtain information. The past researchers focused on the evaluation of influence, based on using the social structure to analyze and find the most influential user. Few researches discussed how the propagation took place between users. And I believe that social network is a heterogeneous network, we cannot just classify a user as a specific topic. User are supposed to have multiple roles in the different topics. With the different influential relationship, user’s identity may be active or passive. So, I propose an integrated analysis system, which supports text-based content and find the valuable feature of social network. The research is composed of four parts. First, I define the hierarchical interesting topic structure generated by data related with targeted users. Second, I try to determine the feature of users according to their activities. User may belong to one interesting topic or many interesting topic at the same time. Third, I extract the influential propagation paths for analysis. The last part is that I build TIPAS (Text-based Influence Propagation Analysis System), which provides analyst to analyze influence which caused propagation paths. For easy understanding, I also use the concept of data visualization to display the analysis results.
