Although social media has become a central arena for deliberative discourse, debates are frequently characterized by incivility. Due to the volume of online discourse, it is challenging to assess when and where interventions, such as community-based counterspeech, are required. Accordingly, automated measurement of deliberative quality is a powerful tool for facilitating constructive online communication. Since deliberation is multidimensional, attempts to monitor online discourse must consider alternative forms of communication to foster inclusive deliberation. Whereas Type I deliberative dimensions, such as rationality and civility, are grounded in Habermasian ideals, Type II deliberation extends this perspective by incorporating alternative communicative forms, including emotionality and storytelling, into a more inclusive assessment of deliberative quality. Addressing the limitations of prior studies that primarily focused on classic Type I deliberative dimensions and tested only a small number of large language models, this study evaluates multiple open-weight models to conduct a nuanced assessment of deliberative quality. We applied the large language model that achieved the best overall balance between predictive performance, model size, and computational efficiency to a dataset of 64,223 comments from German-language climate change discourse on X. Our results demonstrate that large language models can reliably capture complex Type II deliberative dimensions despite class imbalances and subtle linguistic cues that generally hinder automated text classification. Class imbalance often impairs classification performance due to highly skewed label distributions, and Type II dimensions are particularly difficult to detect as they are often signaled by implicit, context-dependent markers. The empirical analysis also reveals distinct communicative patterns: While public speakers predominantly employ civil, rational, constructive, and common-good-oriented communication, other debating users exhibit higher levels of reciprocal engagement, humor, and negative emotions, alongside more frequent incivility. Our study demonstrates that AI-driven measurement can serve as a foundational tool for identifying norm violations, directing counterspeech toward specific debates, promoting inclusive digital public spheres, and observing broader long-term effects of moderation and interventions on the quality of debate.
Political communication research relies on a fragmented landscape of media sources, text and multimedia corpora, analytical tools, and research workflows. Researchers face two related barriers: identifying which resources exist and determining whether and under what conditions they can be accessed and reused. This article presents Monitoring Electoral Democracy (MEDem), a distributed European research infrastructure for comparative political communication research, with particular attention to its Service Center for Text and Multimedia Data. The Service Center develops two complementary services: MEDem Meteor provides a curated, knowledge-graph-based discovery layer that connects media sources, datasets, archives, analytical tools, and publications; MEDem Amsterdam Content Analysis Toolkit supports the hosting, controlled reuse, and non-consumptive analysis of text and multimedia corpora. Together, the services address discovery and access as interconnected infrastructural problems. MEDem Meteor makes dispersed resources, relationships, and coverage gaps visible, while MEDem Amsterdam Content Analysis Toolkit makes access conditions transparent and enables reuse without redistributing protected content. This combination can reduce duplicated data collection, support more theoretically informed source selection, and strengthen cumulative and comparative research. Its realization, however, depends on sustained curation, community participation, and long-term institutional commitment.
The attention-grabbing and memorable nature of visuals makes them a powerful vehicle for spreading hate online. As platforms increasingly reconsider or scale back content moderation efforts, user-centered interventions are gaining importance as complementary governance strategies. While prior research demonstrates the effectiveness of empathy-based textual counter speech, little is known about the role of visual counter speech in mitigating hateful imagery. This preregistered study examines the effectiveness of user-centered interventions against visual hate speech. Using an experimental design (N = 666), we tested four interventions against a no-intervention control condition in response to hateful images targeting transgender persons: (a) empathy-based textual counter speech, (b) AI-generated generic counter images, (c) AI-generated specific counter images replicating elements of the original, and (d) platform-issued warning labels. Results show substantial variation across interventions. Empathy-based textual counter speech and AI-generated specific counter images showed the most consistent reductions in amplification-oriented behaviors, particularly forwarding. Visual interventions did not outperform textual counter speech, suggesting that effectiveness depends less on modality and more on the specificity and contextual alignment of the intervention. Furthermore, effectiveness varied with the intensity of the hate content. Interventions were most effective for humor-based, more ambiguously harmful content, whereas their impact diminished for more severe forms of hate. The consistent performance of AI-generated specific counter images further points to the potential of generative AI to provide users with a rapid, low-cost, and accessible means of producing context-sensitive counter speech.
Artificial intelligence (AI) is increasingly promoted as a transformative force in journalism, yet most research on newsroom AI adoption remains concentrated in Western and other well-resourced media settings. This study examines how AI is interpreted, organized, and used in Malaysian newsrooms, offering insight from an emerging Southeast Asian media context. Drawing on semi-structured interviews with 20 journalists across six Malaysian news organizations, the article investigates how national AI governance signals are translated into newsroom norms, how organizational resources and managerial priorities shape uptake, and how AI is incorporated into everyday reporting, editing, and verification. The analysis shows that AI adoption in Malaysian newsrooms is uneven and selective rather than uniform. First, national AI momentum has not yet produced strong newsroom-level governance; instead, the practice remains shaped by limited written rules, selective disclosure, and strong human accountability. Second, organizational capacity is patchy, with uneven access to tools, training, and infrastructural support across organizations and roles. Third, AI is used mainly in bounded, assistive tasks such as transcription, translation, summarization, and headline generation, while verification labor is redistributed rather than reduced. The findings identify a typology of newsroom responses—cautious minimalists, experimental pragmatists, and human-gated adopters—showing that Malaysian newsrooms are not simply “jumping on the AI bandwagon,” but engaging AI pragmatically under uneven governance and resource conditions.
This article examines the use of social media images during the Tigray War (2020–2022), one of the world’s deadliest armed conflicts of the 21st century. Drawing on a corpus of 92.9 million tweets and over 500,000 images shared on Twitter, we analyse what types of images were shared by the opposing sides of the conflict and what specific functions these images served in antagonistic information campaigns. The article shows how images were used not only to amplify the opposing strategic narratives about war but also to shape the emotional or affective registers related to it. The pro-Tigray campaign relied on the heavy use of emotive images of casualties as well as images of material destruction, which, combined with the use of infographics, collectively conveyed victimhood with the goal of putting pressure on the Ethiopian government and its allies. The pro-government side, in turn, relied more on images that tried to evoke patriotism and resistance in the audience, organise protests, as well as represent a stance of anti-colonial defiance against international pressure. The article contributes to the growing literature on digital media in contemporary conflicts by proposing a novel research framework and methodology and providing original empirical findings on an understudied dimension of contemporary conflicts.
