Predicting political affiliation of posts on Facebook
Che-Chia Chang, Shu-I Chiu, Kuo-Wei Hsu · 2017
Recently, social media such as Facebook has been more popular. Receiving information from Facebook and generating or spreading information on Facebook every day has become a general lifestyle. This new information-exchanging platform contains a lot of meaningful messages including users' emotions and preferences. Using messages on Facebook or in general social media to predict the election result and political affiliation has been a trend. In Taiwan, for example, almost every politician tries to have public opinion polls by using social media; almost every politician has his or her own fan page on Facebook, and so do the parties. We make an effort to predict to what party, DPP or KMT, two major parties in Taiwan, a post would be related or affiliated. We design features and models for the prediction, and we evaluate as well as compare them with the data collected from several political fan pages on Facebook. The results show that we can obtain accuracy higher than 90% when the text and interaction features are used with a nearest neighbor classifier.