With news being consumed more and more online and individually, news recommender systems built into current platforms such as Facebook have a growing influence on the set of news we get to see and consume. This gave rise to a debate whether filter bubbles can occur, i.e., system states in which individuals get to see only those news that support (and amplify) their opinion, which may ultimately lead to opinion polarization or even radicalization. In this project, we will conduct a number of controlled experiments to analyze and measure the effect of such filter bubbles. Given those results, we will derive guidelines for developing responsible news recommender systems which avoid the emergence of filter bubbles, and build proof of concept implementations to analyze the effectiveness of those guidelines.
In this project, FIZ ISE will test and develop automatic text and pattern recognition methods using machine learning, which will be applied to a selected digitized document stock to determine its potential for indexing and online publication.