線上支持度與線下選舉結果之關連性研究

Abstract

隨著網際網路與資訊科技的快速發展之下，在線上則出現了愈來愈多的社群平台以及新聞平台，使得使用者能夠以簡單快速之方法獲取最新消息，此外，因應社群媒體之興起，有愈來愈多的網路使用者會藉由線上社群平台即時分享或發布個人資訊以及各種言論，時至今日，不僅是商業應用亦或是競選團隊都開始重視線上社群平台對於線上網友的經營，以獲取線上網友的支持，然而使用者對於各線上平台的使用態度不盡相同，故本研究希望提出一個有效的方法來針對選舉議題進行探討，藉此找出多個線上平台對於選舉真實結果的關連性。 本研究以時間序列的方法為基礎，針對線上平台之時間序列資料集進行分析，除線上討論熱度之外，更加入情感分析之因素，期望藉此能有效預測與分析於選舉議題的相關文本上，並評估與線下真實結果較具關連性之平台以供未來參考。 本研究針對臺灣最常使用之兩大主流社群平台 Facebook、PTT (包含政黑板、八卦板) 以及最熱門之三大線上新聞平台，蘋果日報、中時電子報與聯合新聞網之資料進行蒐集，並以每日為單位來建置選舉議題之時間序列，最後從五個平台、六種資料集的實驗結果中證明時間序列與情感分析之方法能夠對於線上討論參與選舉之候選人的熱度以及具有情感層面的支持度有良好的預測結果，並且在最後評估時亦證實了線上平台的支持度有一定的程度能夠反映線下選舉結果，而這些發現則能夠使參選人的競選團隊可以針對未來結果與議題之發展有所瞭解並在將來能夠提供更低成本、快速且準確之預測。

With the rapid development of the Internet and the information technology, there are more and more online social platforms and online news platforms. In addition, the users are able to get the latest information and news by a simple and quick way. Moreover, there are an increasing number of Internet users to share or to publish personal information and statements instantly via online social platform. At present, not only the commercial corporation but also the campaign team of election starts to focus on the online social platform and the engagement with the Internet users in order to get their support. However, each user’s attitude of using online platforms is different, so the study is to propose an effective way and find out the relations of online candidate popularity and offline election outcome. In the study, we analyze data of online platform based on the method of time series. Besides, we also consider the factor of sentiment and apply sentiment analysis. We expect that we can forecast the election issues in an effective way and evaluate the relations of online candidate popularity and offline election outcome for reference in the future. We find out there is a correspondence between the share of online popularity and the share of votes in the elections with the method of time series. What’s more, there is an effect of improving our result by considering the factor of sentiment. These findings will be helpful for the election campaign team to have a good understanding of the issues in a fast and lower-cost way.