Unraveling Multidimensional Digital Threats to Democracy: A Case Study of the Austrian Corona Protest Movement's Digital Mobilization

In recent decades, digital platforms have evolved into dominant infrastructures for political communication, presenting both opportunities and threats to democracy (Lorenz-Spreen et al. 2023). The impact of political communication on digital platforms is intricate, involving interactions among actors, technological possibilities, and diverse audiences, and may be heterogeneous and conflicting; not only between problematic and democracy-fostering developments but also within them. Existing empirical research has not fully embraced these multidimensional interactions. Studies tend to isolate investigations to individual platforms (Golovchenko et al. 2020), treat digital threats as independent (McKay & Tenove 2021), and focus on short-term issues (Miller & Vaccari 2020). This study addresses this gap by adopting Lewis and Westlund’s 4A framework (2015; Westlund 2021), conceptualising the prevalence of digital threats as a complex interplay of communicative activities - selection, production, distribution, and interpretation of content - among actors, technological "actants", and audiences in political online communication. Through a case study on the Austrian Corona protest movement's multiplatform mobilization, we aim to identify multidimensional threats and assess their stability over time. The dataset comprises 79,560 postings from 111 relevant actors across ten social media platforms and three data collection periods. On the content level, machine learning classified postings for references to conspiracy theories, hate speech, or antagonists. Actor actions were analyzed by categories, like production of original contributions, or post distribution through sharing. User actions were examined, including selective self-curation through following specific accounts or platform-specific engagement with postings. On the social media level, postings were related to affordances, like varying levels of searchability or scalability. Finally, 38 categorical variables characterizing postings were transformed into embeddings using a Sentence BERT model (Reimers & Gurevych 2020), reduced in dimensionality, and clustered (McInnes et al. 2017, 2018). This allows for a more precise summary of differences and similarities compared to traditional multivariate categorical data analysis methods. The relationships of clusters with several supplementary variables complement the interpretation, such as the functional roles of the actors or the data collection periods. The 35 identified clusters reveal nuanced democracy-threatening potentials at the intersection of communicative actions. For instance, radical posts, referencing multiple antagonists, conspiracy theories, and hate speech, are predominantly found on limited platforms without algorithmic distribution, requiring conscious user activities for discovery. A stable pattern shows a small number of radical posts efficiently reaching many followers. Conversely, there are many radical posting activities that tend to be spontaneous, distributive and reach only a few followers. Civil society actors exhibit stable patterns, with "inwardly directed" radical messages on platforms with little content moderation and "outwardly directed" moderate content widely distributed via mainstream platforms. This approach provides a dense, multidimensional description of digital threats, detecting not obvious anti-democratic activities by examining interactions across platforms and time. It aids in understanding digital threats to democracy as a phenomenon with varying degrees. Emphasizing the complex interplay of communicative actions, it extends solutions beyond regulation to encompass political, technological, and temporal dimensions. Nuanced examinations are crucial for crafting responses to the multifaceted ecologies of digital threats.