Prevalent Discord. Exploring and estimating the prevalence of the type of user disagreement on news media Facebook posts discussing the Colombian peace process (2020-2022)

Sammanfattning: This thesis is dedicated to exploring and understanding public reactions within negotiated peace settlements based on social media data. Concretely, to modeling public opinion and sentiment within the context of the Colombian peace process using a curated dataset of N= ~1.3 million user comments expressing discord on 15,509 Facebook posts, throughout three years (2020-2022). A critical period embracing unprecedented sociopolitical events such as the COVID-19 health emergency, the waves of the Estallido social and the rise to power of the first leftist president in the country. This information was facilitated thanks to the research initiative Agonistic Algorithms from the PUSHPEACE project at Lund University’s Department of Political Science. Based on specialized literature, predictive modeling with a binary logistic regression strategy was employed to discern if, on aggregate, the user comments on a post enabled by news media entities were predominantly antagonistic or not. This approach considered an array of predictors encompassing linguistic features, temporal indicators, engagement metrics, and contextual elements extracted from the Facebook posts. The results indicate limited explanatory capabilities of the exploratory model. Yet, it performed with moderate predictive accuracy on unseen data (64% of overall correct classifications). Regarding the particular status of prevalent antagonism, the model correctly identified this category in 8 out of 10 cases. The covariate referring to location of the publisher of the post emerged as the most influential factor. Despite the limitations, the results suggest that Bogotá-based post publishers carry a higher likelihood of eliciting prevalent user antagonism in comments, compared to posts enablers from other locations.