Competing feedback channels drive phase transitions in collective emotion

Collective emotion on social media can appear stable for long periods before shifting abruptly into self-sustaining polarization. Existing opinion-dynamics models explain abrupt shifts through nonlinear response, threshold activation, or network structure, but they usually treat the information environment as a single channel. They therefore say little about how stability changes when stabilizing mainstream media and amplifying user-generated content coexist. Here we develop a mixed-feedback model in which collective emotion is shaped by competition between these two channels. The model shows that polarization does not intensify all at once: engagement intensity rises first, as the moderate middle thins out, making activity a leading indicator of approaching instability. Agent-based simulations show that this pattern persists beyond the mean-field setting. We then examine social-media data from the COVID-19 period and find the same two signatures predicted by the model: periods of elevated activity are followed by larger polarization jumps, and periods with stronger user-generated dominance are more volatile. Together, these results show that collective fragility depends not only on what people say, but also on how platform feedback mixes stabilizing and amplifying channels. Monitoring engagement intensity may therefore provide earlier warning of emotional tipping points than content extremity alone. • Mainstream vs. user-generated media competition drives emotional bifurcation. • The critical channel-balance ratio is derived analytically from mean-field theory. • Direction–intensity decomposition reveals activity as an early-warning signal. • Network simulations confirm the transition under heterogeneous connectivity. • Both predicted signatures appear in large-scale COVID-19 social-media data.