Communication mode and sentiment polarization dynamics in social media

Polarization in social media posts has become a concern. A variety of theoretical mechanisms have been proposed to explain the change in polarization. However, empirical evidence regarding the dynamic pattern of social media sentiment polarization is lacking. We analyze the sentiment trends in over 20 million tweets (from Twitter) related to the US Presidential Election in 2020 using the empirical mode decomposition methods to uncover the time-frequency characteristics and the causal relationship between directed and undirected polarization. Our analysis suggests that undirected posts are generally more positive and less polarized than directed messages. Both short-term fluctuations and long-term trend contribute to undirected polarization, while polarization in directed messages is driven primarily by short-term fluctuation. Further, while directed polarization is affected by undirected polarizations in the short-term cycles, directed polarization has a larger impact on undirected polarization in the long run.