Political interest and tendency prediction from microblog data

Online social networks, and especially the popular microblogging service Twitter have taken to be the epicenter of massive social movements, where users often openly express political tendencies - a trend which has led to making the classification of political tendencies from social shares a research question of interest. In this research, we collect and hand label a small subset of political messages sent during the follow up period of Gezi Park protests that took place in Turkey in 2013. We demonstrate that in order to predict political relevance and tendency, a Chi-square statistic based feature selection approach coupled with Support Vector Machine and Random Forest classifiers yields significant prediction power.