Tracking the Conversation around NATO in the Nordic States Using Machine Learning

Abstract: Online communication is makes up a large part of the public discourse around current events, and increasingly has been the target for disinformation campaigns. Clearly, there is a need for some language-agnostic tool to map how the discourse shifts and evolves. As such, the goal of this analysis was to create a tool to collate posts from Facebook, Instagram, Reddit and Twitter into one dataset for comparison and analysis, then to generate weekly topics based off posts containing the term “NATO” from this dataset. In this way BERTopic was used in an attempt to divide posts relating to “NATO” into these topics and then map them week-by-week, creating connections between weeks based on the overall topic. This period ranged from the buildup of the Ukrainian/Russian conflict (1 November) until 2 June. The analysis covers four Nordic languages; Danish, Finnish, Norwegian and Swedish from the aforementioned platforms. The Swedish model proved unfruitful due to the stochastic nature of the algorithm, however for the Danish, Finnish and Norwegian datasets, broad overall topics were generated as well as individual weekly topics for each of the broader topics.