Automated analysis of the discussion culture in German news media

Automated analysis of the discussion culture in German news media

Institute(s): TU Darmstadt

Associated since: March 2017

CEDIFOR project participants

Prof. Dr. Iryna Gurevych, UKP Lab, TU Darmstadt
Dr. Johannes Daxenberger, UKP Lab, TU Darmstadt

CEDIFOR Partner:

Prof. Dr. Oliver Quiring, Institute of Journalism, JGU Mainz
Dr. Marc Ziegele, Institute of Journalism, JGU Mainz

brief description

The current debate on user and citizen participation on the Internet revolves to a large extent around negative and uncivil excesses. This phenomenon includes, for example, hate comments, the creation of “alternative realities” or the targeted sowing of mistrust in journalism by means of manual (by people) or automated (by social bots) commenting. In this project we are especially dedicated to the commenting of mass media news (e.g. on the websites of tagesschau, Spiegel etc. and their social media presences), because these are the places where particularly heated debates are publicly held.

Unciviliar comments are problematic because they undermine democratic values, can promote stereotypical attitudes of recipients and have negative effects on the perceived news quality of media. Moreover, too much in-civility is diametrically opposed to community building in integrated publics: Many users state that, while they generally perceive online discussions on news sites as social and entertaining events, the aggressiveness and conflictual nature of many of the contributions prevents them from participating. Since a high participation rate of the audience is also desirable from a normative point of view (e.g. audience inclusion) and for media economic considerations (e.g. customer loyalty), the question arises as to how the currently prevailing “mode of conflict” in subsequent communication on news sites can be changed into a “mode of cooperation”.

In this project, the first step will be to develop and test automatic methods for identifying and predicting civil and uncivil discussion processes. In a second step, these methods will be used to identify uncivilized discussion processes at an early stage across all media and on a broad basis using machine learning, and to examine them with regard to their triggering criteria (e.g. message and discussion values). In the medium term, suggestions for improving the establishment of a more civil debate culture will be developed. The evaluations will be carried out in cooperation with our partners from communication science and journalism at the JGU Mainz.

Project results


  • Ziegele, M., Daxenberger, J., Quiring, O., & Gurevych, I. (2018). Developing Automated Measures to Predict Incivility in Public Online Discussions on the Facebook Sites of Established News Media. In Proceedings of the 68th Annual Conference of the International Communication Association (ICA).
  • Daxenberger, J., Ziegele, M., Gurevych, I., Quiring, O. (2018): Automatically Detecting Incivility in Online Discussions of News Media. In: Proceedings of the 14th eScience IEEE International Conference, Amsterdam, Netherlands.