Real-time large-scale data streams provided by Twitter create new possibilities for political scientists to nowcast political events. We developed a pipeline to process, analyze and aggregate data for presentation on a web application. During the 2021 German federal election campaign, expressed stances on competing political parties and their front-runners were analyzed in real time. State-of-the-art linguistic neural networks were reused and adapted by post-training and fine-tuning for detecting stances toward political actors. Furthermore, a dictionary-based approach was adopted to analyze the salient topics during the campaign across 32 (policy) issues. Within stance detection, a decrease in performance over time became visible, which can largely be attributed to a shift in issue focus on Twitter during the election campaign. This is emphasized with concrete empirical examples. During the final phase of the campaign, qualitative monitoring was maintained to ensure the validity and reliability of our findings. Based on this, potential error sources are presented, and possible solutions for future research are offered.    «

Real-time large-scale data streams provided by Twitter create new possibilities for political scientists to nowcast political events. We developed a pipeline to process, analyze and aggregate data for presentation on a web application. During the 2021 German federal election campaign, expressed stances on competing political parties and their front-runners were analyzed in real time. State-of-the-art linguistic neural networks were reused and adapted by post-training and fine-tuning for detectin...    »