多重社群媒體網站事件整合與主題偵測之研究

Abstract

社群媒體網站的興起，加速了媒體訊息的傳播速度，其影響力既深且廣，也興起了新一代的網路應用變革，例如：社群媒體行銷活動的日漸擴大，民主時代的競選活動也已廣泛採納社群媒體與傳統媒體並行的方式。然而，網路大眾所分享的事件與資訊，卻可能大量散佈在不同的「社群媒體網站」之中（如：推特－Twitter、臉書－Facebook，或領英－LinkedIn），若單純只看單一社群媒體來輔助某些決策，便可能有部份的遺珠之憾，若是參與多個社群媒體，卻也可能因為重大事件被重覆散佈而造成資訊過載（Information Overloading）的困擾，加上社群媒體裡的資訊更新速度相當頻繁、類型多樣化、數量龐大，長期以來已經造成社群媒體內容分析的困難度。因此，本研究提出了一個適用於中文環境的社群媒體智慧整合平台，可以從各個不同的社群媒體網站中自動辨識出已知事件的相關文件，並自這些文件中偵測出事件的主題，然後建置主題的圖形資料庫與即時商業智慧分析介面，讓使用者可以用視覺化的方式（如：多維度表格、主題圖形）閱讀整合後的訊息，進一步提供使用者進行關聯追蹤以及趨勢分析的決策服務。

In recent years, the rise of social media sites has further accelerated the spread speed of media information, and the deep influence has also spawned a new paradigm shift of Web applications, such as: the expansion of social media marketing, democratic campaign activities have been widely adopted with the way social media and traditional media work side by side. However, many public event information shared by the community are mostly scattered in various different social media sites (e.g., Twitter, Facebook or LinkedIn). When only a single social media is used to assist in certain decisions, there may be some pity caused by incomplete information. On the other hand, by taking multiple social media sites for reference among the major events, we still suffer from the information overloading problem, as some parts may possibly be repeated by a large number of users across difference sites to spread over and over again. This not only makes the media events be updated faster with various diversity, but also be pasted with noise or redundancy, causing inconvenience on social media analysis. Therefore, this study intend to propose a social media intelligence integration platform for the Chinese environment, which can automatically identify relevant information of the extracted events from different social media websites, and then build the topic graph and real-time business intelligence analysis interface, such that users can visually (e.g., multi-dimensional tables, topic graphs) interpret the message and precede the topic modeling, event tracking, and trend analysis for further decision making.