On Saturday 30 September 2023, Slovakia held early elections to the National Council of the Slovak Republic, which were preceded by an intense election campaign largely taking place on social networks. In cooperation with Demagóg.sk and Infosecurity.sk, the analytical department of the Council for Media Services prepared an exploratory report to analyse the content related to the elections on one of the most widely used content-sharing platforms – Facebook. Statistical programme and statistical topic modelling using unsupervised machine learning were used to analyse election narratives.
Key findings:
In the examined period from 10 June 2023 to 15 October 2023, 12,301 posts related to the early parliamentary elections in Slovakia were recorded on Facebook, based on the predefined conditions and after data cleaning. These received a total of 8,552,967 interactions.
Based on the analysis of word frequency in the selected dataset, 57 most common topics were identified. 16 of these were subjected to in-depth qualitative analysis and described.
Among the qualitatively processed themes, those that objectively informed about the election, such as election polls, voting from abroad, or transparency of electoral processes, emerged.
However, a significant part of the topics focused on challenging or criticizing public institutions, such as the Office of the President, the Specialized Prosecutor General’s Office, the Police Force or the Ministry of Interior. Finally, they included topics dealing with geopolitical issues, such as Hungary's influence on the election campaign or the war in Ukraine, as well as reporting on the activities of political parties and individuals.
The report also examines the most viral posts of the examined period, albeit to a limited extent. Eight of the top ten most viral posts belonged to the chairman of SMER-SSD, Robert Fico. They generated 257,414 interactions in total. Seven of them were accompanied by video recordings, which generated a total of 2,869,120 views.
The analysis shows that the most prevalent topics (i.e. quantitatively most recurring) were not automatically the most viral ones (i.e. posts with the quantitatively highest number of interactions). However, some topics could be classified as both prevalent and viral.
An analysis based on statistical software and statistical modelling using unsupervised machine learning proves to be a suitable tool to gain a better understanding of the characteristics and nature of the information environment. Indeed, the results suggest that it is not only the virality of posts which requires examination but also their prevalence.
The exercise detected a significant amount of content produced by individuals whose accounts were suspended due to repeated violations of Facebook’s terms of use. This raises the question whether the measures for repeated violations of the platform’s rules are effectively and appropriately enforced.