運用線上社群資訊於使用者行為建模與預測

隨著線上社群網路的快速發展，大量使用者產生的內容在網路上流通。在此博士論文中，我們運用了大規模的線上社群資訊，提出一系列的技術用來預測群眾回應、事件結果及群眾對於特定議題持有的態度。首先我們提出了一套能夠迅速偵測社群重要文章的方法，包含被大眾廣泛注意並回應，以及獲得一面倒支持或反對的文章。此外，我們也建構一套偵測具影響力作者的方法，此方法能找出有效獲得群眾廣泛回應並取得大量使用者讚許或反對的文章作者，我們認為這些作者將會是在線上具有影響力的資訊傳播媒介。透過我們利用社會運動與選舉活動時的線上討論平台實證研究，我們的方法能有效在短時間內偵測重要文章並分辨出具有影響力的作者。接續，我們專注於應用社群資訊來預測社會事件結果，我們選擇選舉預測來作為一個應用的場域。在研究中我們結合了情緒分析與使用者行為研究來建構出一套選舉得票率與勝負預測模型。我們利用 2014 年台灣六都選舉來進行驗證，結果顯示我們的方法較先前的方法更加精準，不管在得票率或勝敗預測都取得較佳的結果。最終，我們延伸了利用社群媒體進行研究的議題至大規模的 Facebook 粉絲頁政治傾向分析，此研究目的為找出潛在的政治討論區與宣傳媒介。我們針對台灣約 20 萬 Facebook 粉絲頁進行其政治傾向與政治參與度進行研究。在研究中我們提出了對於不同類別的粉絲頁之政治傾向分數與參與度估計的方法，同時也提出了一套可以判別攻擊性與反諷性政治粉絲頁的偵測機制。

關鍵字

社群媒體分析；資訊過濾；選舉預測；臉書；政治傾向

並列摘要

With the rapid development of online social networks, increasing amounts of user-generated content are posted online. In this dissertation, we aim to leverage the large-scale online information to propose techniques for forecasting the public responses, the event results, and even the attitude of human beings toward a certain issue. First, we develop methodologies which identify the important articles from two perspectives: frequent discussion and extreme acceptance or rejection by the online public. In addition, we discuss approaches for distinguishing influential authors who are popular and receive consistently high ratings from online users. Our results demonstrate that our methodologies achieve high accuracy with significant time reduction and outperform previous methodologies in distinguishing notable articles and authors on social forums. Second, we aim to use the social context to predict the results of events. We select election prediction as an application of the prediction methodology. We incorporate sentiment analysis with user behavior investigation to establish the prediction model. We validate our methodology in the 2014 Taiwan elections, and our proposal outperforms previous approaches in predicting the final winners of the elections as well as the final vote shares. Last but not least, we further investigate the potential online channels for political information dissemination and for politicians to broaden their connections with users. In the research we examine the ideological score of more than 0.2 million Facebook pages in Taiwan, where politics has long been polarized between the two major parties the Kuomintang (KMT) and the Democratic Progressive Party (DPP). From our approaches, we estimate the ideological scores of different kinds Facebook pages. Our contributions including an empirical political ideology estimation of pages with different perspectives (politicalness and category), but also applications of identifying potential political and conflicting/sarcastic pages.

並列關鍵字

Social media analysis ； Information filtering ； Election prediction ； Facebook ； Political ideology

參考文獻

[1] E. Bakshy, S. Messing, and L. A. Adamic, “Exposure to ideologically diverse news and opinion on Facebook,” Science, vol. 348, no. 6239, pp. 1130–1132, 2015.

[2] P. Barberá, J. T. Jost, J. Nagler, J. A. Tucker, and R. Bonneau, “Tweeting from left to right is online political communication more than an echo chamber?” Psychological Science, vol. 26, no. 10, pp. 1531–1542, 2015.

[4] S. J. Barnes and A. D. Pressey, “In search of the “meta-maven” : An examination of market maven behavior across real-life, web, and virtual world marketing channels,” Psychology & Marketing, vol. 29, no. 3, pp. 167–185, 2012.

[5] J. N. Bassili, “Response latency and the accessibility of voting intentions: What contributes to accessibility and how it affects vote choice,” Personality and social psychology bulletin, vol. 21, no. 7, pp. 686–695, 1995.

[6] J. E. Berg, F. D. Nelson, and T. A. Rietz, “Prediction market accuracy in the long run,” International Journal of Forecasting, vol. 24, no. 2, pp. 285–300, 2008.

國際替代計量

運用線上社群資訊於使用者行為建模與預測

全文下載

主題瀏覽