The rise of populism in the past decades is a political trend, which binds the attention of political scientists around the world. However, the tools for measuring populism, or populist speech, still bear improvement. In order to achieve such improvement, the use of data from social media is particularly promising. Firstly, it can be assumed, that populist rhetoric is particularly present in media with few gatekeepers. Secondly, the data available in social media is highly accessible. By reason of the particular suitability of social media data, we used such data, for testing approaches to classifying populist speech. These approaches include classifications based on: predefined keywords, computed keywords, classification trees and regression. The data used consists of Facebook posts and related user comments published during the German federal election of 2017 - the first election in decades, that resulted in mandates for a German right-wing populist party. Training and test datasets have been coded by well-trained coders according to a resemblance of different types of populism (nationalism, anti-establishment). The results of the analysis are meant to provide researchers with a reliable procedure for identifying populist rhetoric in social media.
