Public Opinion Toward CSSTA: A Text Mining Approach

Recently, the public of Taiwan has had a heated debate on the issue of Cross-Strait Service Trade Agreement (CSSTA). After months of simmering tensions between ruling party and opposition party strongly backed by the student-led Sunflower Movement, the debate has finally reached a breaking point on March 18, 2014, at which students occupied the Legislative Yuan. During this period, novel communication such as Facebook sharing, instant messaging, and discussions on PTT have reshaped the social movement since they are easily accessible and instantly responded. The social media has become the dominant source in opinion shaping and the accompanying sentiment spread. The extraction and tracking of uprising political opinions and events such as CSSTA has become one of the most important topics that receive much attention. With the huge amounts of texts, it is not possible to analyze and interpreting the social and political texts manually. Instead, we propose to use the text mining approach, which automatically extract opinion and information profiles from the texts. Moreover, this approach also strengthens the objectivity, for the norms are set a priori, and thus human biases are reduced. As a pioneering work in the context of Taiwan society, this research aims to trace the public opinion toward CSSTA from the perspective of text mining. The approach involves the manually extracting of political stance related keywords and phrases, supervised machining learning, and a statistical model of the trend. We focus on the individual posts on PTT rather than news since they are more representative. The potential political or commercial applications are valuable. One can discover the public opinion and response in a short time. The materials we used in this research includes a list of manually created seed words and phrases representing the pro-and-con political polarity, respectively. These words are tested by the texts of the website “服貿東西軍”, which classifies the supporting and opposing texts. Another resource we used in this work is the PTT corpus, which is a popular online bulletin board favored by many of the youth. * Graduate Institute of Linguistics, National Taiwan University † http://ecfa.speaking.tw/imho.php