Machine learning for cross-platform political communication research: Argentine government and opposition in Facebook, Instagram and Twitter

| This article studies political communication in different platforms, applying data science methods to analyze similarities and differences among Facebook, Instagram, and Twitter posts of 50 Argentinian politicians in 2020. This is a pioneering cross-platform study for our region, and its objectives are heuristic and methodological. Regarding the former, we show that strategies differ among platforms: Twitter is the battlefield for controversy and interpellations among politicians, and toxicity is rewarded, while on Facebook and Instagram politicians expand on the topics in which they seem to consider themselves stronger. The closs-platform study shows that even in a polarized context such as the Argentinean one, there are common and non-controversial topics. Methodologically, we use novel analytical methods and implemented a recent topic-detection algorithm, we apply sentiment analysis techniques to understand if texts have positive or negative intentions, and deep neural networks to detect toxicity in a text, among others. Readers are offered access to the toolbox developed during the research, which can be useful for working large text corpora.