Marc Ziegele

This study tests whether an AI-based counter-attitudinal comment recommender (AI-CAR) enhances participants’ subjective deliberative experiences and reduces incivility by restructuring the visibility of dissenting viewpoints. We analyzed three-wave data from 827 participants in a 10-day simulated digital citizens’ council in Germany. Participants discussed one of two political issues asynchronously and were randomly assigned to one of three conditions: counter-attitudinal AI recommendations, randomly selected recommendations, or no recommendations. Assignment to AI-CAR increased perceived equality of viewpoints. Higher perceived equality at wave 2 was associated with greater internal deliberation and discussion satisfaction at wave 3, controlling for their wave-2 levels. The corresponding model-based indirect effects were statistically significant, whereas comparable indirect effects were not statistically significant for randomly selected recommendations. AI-CAR did not reduce incivility in participants’ contributions. Overall, the findings highlight the potential of AI-driven structural curation to support subjective deliberative experiences, while effects on incivility remain limited.

Political communication scholarship on citizen participation has mostly focused on the exercise of democratic voice (e.g., Theocharis & Van Deth, 2018). Especially the Internet has been hailed for ...

This chapter provides an overview of the work of the junior research group DEDIS (Deliberative Discussions in the Social Web). Drawing on theories of deliberative democracy, we developed and tested innovative measures to improve the quality of citizens’ online discussions about political issues. This chapter focuses on our research on new forms of moderating online discussions and our development of computer-based moderation systems. It outlines our model of moderation effects, which suggests that moderation increases citizens’ willingness to participate in online discussions and to write engaging comments through two pathways: perceived support/sense of community and perceived discussion quality. The chapter also describes empirical evidence that such an expanded understanding of moderation and its effects improves both the quality and deliberative outcomes of online discussions. Finally, the chapter includes a brief overview of an AI-powered moderation system we developed that can help moderators efficiently identify and respond to high-quality and low-quality user comments. Overall, the work of the junior research group advances research on online deliberation and moderation and helps to formulate realistic expectations about what professional moderation can achieve.

Machine learning models used in algorithmic content moderation in online discussions play an increasingly important role in detecting high-quality (e.g., constructive or engaging) and uncivil (e.g., toxic, hateful, or abusive) comments. Trained on human decisions, these models risk reproducing annotator bias, particularly when the perception of such nuanced categories is influenced by social backgrounds and training data lacks sufficient representational diversity. This study investigates the potential impact of annotators’ formal education on the classification of high-quality and uncivil user comments in online discussions. Using a dataset of 13,677 German user comments annotated by crowd workers with low, medium, and high formal education, we investigate divergences in classification outcomes when classifiers are trained on data annotated by different educational groups. In addition, we assessed statistical associations between fine-grained comment features and the annotations provided by each subgroup. Our results indicate that classifiers for high-quality and uncivil comments yield significantly different results when trained on data labeled by annotators with varying educational backgrounds. Furthermore, comment features used in traditional operationalizations of deliberative quality (e.g., solution proposals, additional knowledge) and incivility (e.g., vulgarity, accusation of lying) are more strongly associated with annotations by crowd annotators with medium and high formal education. We argue that such group-specific differences should be considered when developing machine learning models in both practice and research to reflect a more inclusive understanding of valuable and norm-violating user contributions to online discussions.

Maike Behrendt, Stefan Sylvius Wagner, Mira Warne, Jana Leonie Peters, Marc Ziegele, Stefan Harmeling. Proceedings of the Fourth Workshop on Bridging Human-Computer Interaction and Natural Language Processing (HCI+NLP). 2025.

In unserem Beitrag untersuchen wir am Beispiel der Identitären Bewegung die Auswirkungen von Deplatforming – dem Sperren von Akteuren auf sozialen Netzwerken – auf die Online-Kommunikation von politisch rechten bzw. rechtsextremen Akteuren. Mittels quantitativer Inhaltsanalyse wurde die Kommunikation von zwei Akteuren der Identitären Bewegung auf Telegram vor und nach der Sperrung dieser Akteure auf Twitter untersucht. Vergleiche zwischen den Zeiträumen vor und nach der Sperrung erfolgten hinsichtlich der Kommunikationsfrequenz, der Postreichweite, des Kommunikationsstils und der inhaltlichen Radikalität. Zugrunde liegen der aktuelle Forschungsstand in diesem Themenfeld sowie Theorien von Repressionen, der Gegenöffentlichkeit und des Konzeptes der Dangerous Speech. Es zeigt sich, dass der Kommunikationsumfang nach der Sperrung abnimmt, die Reichweite des Telegramkanals wächst, der kommunikative Stil weitgehend konstant bleibt und eine inhaltliche Radikalisierung nur partiell zu beobachten ist. In our paper, we use the Identitarian Movement as an example to investigate the effects of deplatforming – blocking actors on social networks – on the online communication of politically right-wing or far-right actors. Using quantitative content analysis, the communication of two actors of the Identitarian Movement on Telegram was examined before and after the blocking of these actors on Twitter. Comparisons were made between the pre- and post-blocking periods in terms of communication frequency, post reach, communication style, and radicalness of the content. Our analysis is based on the current state of research in this topic area as well as on theories of repression, counter-publics, and the concept of Dangerous Speech. Our results show that the frequency of the actors' communication decreased after the blocking. The reach of the telegram channels increased, the communicative style remained largely constant, and a radicalization of content could only be observed partially.