A Data Analysis of Election Speeches
This article was probably one of my largest projects at Der Spiegel so far. To analyze the contents of the campaign speeches by the candidates, we implemented a pipeline that classified each sentence into a topic and an emotion by using the GPT-4o modell. To ensure the quality of the results, we used multiple approaches such as classifying each sentence ten times to detect anomalies, returning a likelihood per category between 0 and 1, and controls through handcoding.
The results were then visualized in a design that contained multiple aspects. We showed the topics of each sentence through color and the emotion through the form of the line, underscoring the sentence. We start out by highlighting certain sentences, through cards.
We then guided the readers through the main results of our analysis through a scroller. We highlight certain topics that defined the campaign and explain how the visualization works. With these skills the users can then explore the speeches of the candidates in detail themselves.
For this project, I mostly took care of the data analysis together with Holger Dambeck. The design is by Nina Krug and Helen Bielawa. Helen also implemented most of the Frontend, I helped out with this a little bit. Janine Ahrendt and Holger Dambeck wrote the texts.