半結構化語意模型分析技術之開發-以社群網路資料為例

隨著資訊科技的日新月異以及網際網路應用的興盛，網路上所提供的服務已廣泛應用於我們日常生活工作當中，且社群網站的發展為人們帶來了變化，改變了人們的交際模式，也強烈的影響人們的生活。現今Hito商品或熱門話題於Facebook上都有其專屬的社團以及粉絲團，版上常會公佈新產品動向或是關於產品的評論、討論文。Facebook晉升為新一代熱門交流平台，Facebook社群中的資訊內容是經由每位使用者的參與而產出內容，經由人和人之間的分享，造就了現今的社群網路知識庫。如果使用者想快速從這龐大的知識庫中，取出自已有興趣的訊息和了解目前最受討論的熱門主題情形，在過去的情況，得需依靠人力做市場口碑調查，以時間換取效益的方式。此一方式雖然可行，但傳統口碑調查多仰賴口耳相傳，不易保存且可信度也有限。相較於網路口碑，因為其資料儲存於網路中，造成的影響力遠大於傳統口碑，網路上的口碑討論已是影響消費者購買決策的重要因素。近幾年，已有許多研究針對商務網站的評論分析網路口碑的影響，但卻少有研究綜合性社群網站的評論分析來做為網路口碑的推斷依據。如何用最快速度幫使用者篩選出精準資訊，就是未來能勝出的人。但是，網路上的留言與新聞公告、電子郵件、技術文件、會議記錄等都是屬於非結構化的資料，必須透過文字探勘(Text mining)的技術，才得以發掘隱含在文件中的知識。因此本研究試圖提出具體可行的架構，結合Facebook應用程式，彙整使用者的社團與粉絲團在近月份的討論串至資料庫，進而透過語意拆解與分析取出有價值的討論串，再與事實結果比對進行驗證，提供分析資訊作為決策的依據。

關鍵字

半結構化資料；語意分析；社群網站

並列摘要

With the advancement of information technology and the flourish of internet applications, the services which provided through by internet have been widely used in our daily life and work. The development of social networking sites brings change for people. They not only change people's communication mode, but also have a strong impact on people's lives. Today, the Hito good or a hot topic on Facebook has its own club and the fans group. The clubs often publish the trend of new product or the comments and discussion about the product. Therefore, Facebook promote as a new generation of popular communication platform. Each output information from Facebook’s content is because of each user’s participation. And these information by sharing between people, the knowledge base of social network is created nowadays. In the past, if a user wants to select accurate information out fast from a large knowledge base, it is necessary to need human doing the survey of market reputation. But the traditional reputation surveys rely on word of mouth. While they are more difficult to preserve and have the limitation of credibility. On the other hand, compared to internet reputation, its data is stored in the network, causing much greater than the influence of the traditional reputation, even an important factor for purchasing decisions. However, they are unstructured data between comments, emails, technical documents and meeting minutes. These must use the way of text mining to uncover the knowledge that implied in the document. The study attempts to make a feasible framework and combine with the Facebook application. Compiling the discussions string to database and revealing the valuable one by semantic analysis. Then compared with the result of the fact to verify to be a basis for decision-making.