Measuring partisan community dynamics: a longitudinal analysis of affective engagement in pro-Bolsonaro Facebook networks

This study examines how major political events shape affective engagement patterns in a hyperpartisan pro-Bolsonaro network of Facebook accounts. Hyperpartisan communities are often characterized as impermeable echo chambers with predictable engagement patterns, yet this assumption has rarely been tested longitudinally. Using a mixed-methods approach combining LLM-based content classification, time series analysis, and regression modeling, we analyze over 12 million posts from 53 Brazilian pro-Bolsonaro groups and pages across 2021–2023. Two behavioral indices were constructed: the Emotional Polarization Index measuring love-versus-angry reactions, and the Engagement Balance Index capturing comment-versus-share interactions. Results show that hyperpartisan communities displayed surprisingly unstable engagement patterns over time, contradicting traditional theories that assume polarization as consistent. A major shift occurred in 2023 around Lula's inauguration and the January coup attempt, marked by declining emotional intensity, increased internal debate, and weakening opposition responses. These findings suggest that engagement dynamics extend beyond fixed ideological positions, shaped by major political events and contextual changes. The study contributes methodologically through its longitudinal approach and use of machine learning for political actor recognition, offering new insights into the dynamic nature of digital political polarization.