Stance detection holds great potential to improve online political discussions through its deployment in discussion platforms for purposes such as content moderation, topic summarisation or to facilitate more balanced discussions. Typically, transformer-based models are employed directly for stance detection, requiring vast amounts of data. However, the wide variety of debate topics in online political discussions makes data collection particularly challenging. LLMs have revived stance detection, but their online deployment in online political discussions faces challenges like inconsistent outputs, biases, and vulnerability to adversarial attacks. We show how LLM-generated synthetic data can improve stance detection for online political discussions by using reliable traditional stance detection models for online deployment, while leveraging the text generation capabilities of LLMs for synthetic data generation in a secure offline environment. To achieve this, (i) we generate synthetic data for specific debate questions by prompting a Mistral-7B model and show that fine-tuning with the generated synthetic data can substantially improve the performance of stance detection, while remaining interpretable and aligned with real world data. (ii) Using the synthetic data as a reference, we can improve performance even further by identifying the most informative samples in an unlabelled dataset, i.e., those samples which the stance detection model is most uncertain about and can benefit from the most. By fine-tuning with both synthetic data and the most informative samples, we surpass the performance of the baseline model that is fine-tuned on all true labels, while labelling considerably less data.

Many online discussion providers consider using algorithm-based moderation software to support their employees in moderating toxic communication. Such technology is also attractive for public discussion providers, including public administration and public service media. To ensure successful implementation, however, it is crucial that moderators can correctly understand and use the software according to context-specific workplace requirements. This exploratory case study sheds light on the technology acceptance of algorithm-based moderation software by moderators in German public administration and public service media. Specifically, we focus on the moderators’ user characteristics and workplace requirement perceptions as preconditions for technology acceptance. We combined twelve structured qualitative interviews with moderators with an enhanced cognitive walkthrough (ECW) of an algorithm-based moderation dashboard. Additionally, stimuli of two different transparency mechanisms were added to the interview. Our findings suggest that transparency is one of the most requested characteristics of algorithm-based moderation software and, when met, transparency is beneficial for the acceptance of automated content classification in these systems. However, the findings suggest that different AI perceptions and technology commitment among moderators corresponded with different transparency motives related to the moderation system. We assume that addressing those differing motives by different transparency mechanisms may positively affect technology acceptance.

Detecting biased language in written discourse is a highly relevant area of research in political communication and other social sciences, given the large quantity of information exchanged in public online platforms. In this abstract, we discuss an approach based on the concept of "incivility"-assessing biased text on the Facebook pages of established news media. News outlets are forced to put increasing efforts into preventing heated debates from turning into disrespectful discussions on their social media platforms. By scaling the analysis from a few thousand manually coded samples to more than a million comments, we take a step towards supporting media outlets in (semi-)automatizing the detection of uncivil comments and enable a much broader analysis of the latter.