Coptwitter

During my Bachelor’s at the University of Leipzig, I carried out an explorative data analysis using Twitter data from German police stations. Analysing the content (topic modelling) and tone (sentiment analysis) provided interesting insights into the social media strategies of the German police.

Topic modelling allows you to analyse the content of large bodies of text by means of finding patterns of co-occuring words. Here are the resulting topics and their associated words. Mean theta indicates the average probability that a random tweet will be assigned to a topic.

This plot shows how the probabilities that specific topics emerge in the tweets developed over time. Topic 4 is on corona demonstrations (blue). Topic 9 (red) covers advice to mind iced out streets. Topic 13 (green) covers tweets with advice to take care from tricksters

Topic 10 signals that the police sent out wishes during the advent season. Comparing the pattern with a subsequent sentiment analysis (not free from methodological bias), it shows that these tweets also seem to be the most positive in tone.

Resources

Find the dataset and more results on my GitHub.