Sie sind hier

Applying AutoML on Political Communication in Twitter Data

We have collected a large set of tweets observing the German federal election in 2017. Tweets originate from candidates to the German parliament, from different media accounts and from people who tweeted using a prespecified set of hashtags. Some of the account holders are described by metadata specifying, e.g. their party or their candidate status. From a political science perspective it would be interesting to learn about commonalities and differences of various candidate and media types. The issue is that a large number of correlations might occur and at priori it is not clear which ones are more interesting than others. AutoML is an approach in machine learning that explores the search space of machine learning models preferring some models over others. It is the task of this master thesis to specify the above domain analysis problem in terms of an AutoML problem and investigate to which extent AutoML may be helpful in finding valid hypotheses interesting to political science. Our political science partners in the project E-Democracy will be available (to limited extent) in order to judge interestingness of resulting models.