以資料探勘方法探究 FACEBOOK 政治人物粉絲專頁網民角色分析

Abstract

本研究嘗試開發出一套可快速分析粉絲專頁中粉絲立場的系統。本論文以三位政治人物粉絲專頁為例，透過Facebook API撈取粉絲留言，並建置支持與不支持的詞彙列表，試圖快速地篩選出在一粉絲專頁中支持與不支持的留言粉絲與其留言。建置詞彙列表的方法為，透過人工標記判斷一留言為支持或不支持，並從該留言中擷取可作為詞彙列表的詞彙，評估方法為觀察準確率與召回率之變化。本研究透過抽樣詞彙列表判斷的結過給予人工進行標記，以來判斷該系統之品質。 實驗結果顯示，在支持方面，本研究所建立的詞彙庫能夠判斷出97.57%的朱立倫支持粉絲數，最少的是蔡英文粉絲團，僅能夠判斷出67.41%的支持粉絲數；在不支持粉絲數方面，則是蔡英文的最多，能夠偵測出32.58%，僅能偵測出朱立倫粉絲中的2.43%為不支持粉絲。比起支持詞彙列表，不支持詞彙列表的建置過程較為困難，原因為表達不支持的方式與詞彙較多，較不能如同支持詞彙般明確且較為固定。另外，不支持的留言往往會運用反諷的方式，留言中雖有支持的詞彙出現，但其語句意思實為不支持，因此提高了不支持詞彙列表增建的困難度。最後，本研究依據研究結果進行討論和提出未來研究相關建議。This research aims to develop an agile system that can screen out the stands of fans who commented under the post of Facebook fanpage fast. The system contains data collection using Facebook API, data storage in MySQL and process of building supportive and non-supportive keyword lists. This research takes three politicians Facebook fanpages for example, by building up keyword lists from the result of annotation, we can quickly filter out the stands of fans and comments. This research uses recall and precision rate to evaluate the quality of keyword lists, and uses annotation to evaluate the quality of this system. The result shows that Chu’s supportive keyword lists perform the best, and Tsai’s supportive keyword lists are the opposite. But when it comes to non-supportive keyword lists, Tsai’s keyword list outstands than others. We can refer that the building process of non-supportive keyword list is much more difficult than building up a supportive one. The reasons are the use of ironic wording and the way to express non-supportive stand varies. Advice is given for the reference of future studies.