Determining the Function of Political Tweets

Erik Tjong Kim Sang, Herbert Kruitbosch, Marcel Broersma, Marc Esteve Del Valle

Abstract: We study the discursive practices of politicians and journalists on social media. For this we need more annotated data than we currently have but the annotation process is time-consuming and costly. In this paper we examine machine learning methods for automatically annotating unseen tweetsbased on a small set of manually annotated tweets. Forimproving the performance of the learner, we focus onmethods related to training data expansion, like artificialtraining data, active learning and incorporating languagemodels developed from unannotated text.