Automatische Erkennung von politischen Trends mit Twitter - brauchen wir Meinungsumfragen noch?

Abstract

Opinion research institutes try to capture the public opinion on politicians with telephone and road surveys. In winter 2015/16, a student group worked on the question if it is possible to automate this process. The idea behind this is that the platform Twitter is often used for political discussions. Since tweets are limited to 140 characters, and the topic can usually be clearly identified by hashtags, Twitter data seem to be well suited for automatic sentiment analysis. Sentiment analysis methods aim to automatically classify these tweets into positive and negative expressions of opinion. We have implemented a Twitter crawler and sentiment analysis routines in Python. We then collected tweets for politicians over a period of four weeks, and visualized the results of the opinion analyses. Finally, we compared our results with the "ZDF-Politbarometer", a monthly TV opinion analysis.