In our first approach we wanted to use VADER¹ for the sentiment analysis because it is said to be good for social media texts. However, because of translation issues we decided to use textblob² which already comes with a german language extension. So we are able to get the sentiment, the polarity and the subjectivity of the tweets.

We decided to identify the topics in the tweets through the LDA probability model. With the library gensim³ we are able to generate a LDA-Model. By varying the amount of topics and comparing the resulting coherence value, we determine the optimal number of topics. With the library pyLDAvis⁴ the topics can be visualized and clustered. Our aims are for example to demonstrate changes over the time or dominant topics of parties.

Follower networks can be represnted by nodes (Twitter accounts) and edges (follower relations between the accounts). networkx⁵ is a Python library suited for this kind of data modelling. We plan to analyze the network created with our collected data further, but for the time being we simply created a first graph showing (very poorly) the relations between the Twitter accounts. The colored dots represent a account (color signifying party affiliation), while the arrows show the follower relations between the accounts.

Additionally, we want to associate COVID-data (like case numbers, deaths, etc) with the twitter data. For this purpose, we will use data provided by OurWorldInData.org

¹https://pypi.org/project/vader-sentiment/

²https://pypi.org/project/textblob-de/

³https://pypi.org/project/gensim/

⁴https://pypi.org/project/pyLDAvis/

⁵https://pypi.org/project/networkx/