Dynamic Emotions of Supporters and Opponents of Anti-racism Movement from George Floyd Protests

TL;DR

This paper analyzes social media discussions during the George Floyd protests, revealing stance shifts over time and challenging assumptions about sentiment and stance prediction, with novel applications of aspect-based sentiment analysis and large-scale stance prediction.

Contribution

It introduces the application of aspect-based sentiment analysis in social science and pioneers large-scale stance prediction during social movements.

Findings

More users shifted stance than maintained it.

Negative sentiment was prevalent across both supporters and opponents.

Sentiment alone is insufficient for stance prediction.

Abstract

Social media empowers citizens to raise the voice and expressed civil outrage leads to collective action to change the society. Since social media welcomes anyone regardless of the political ideology or perspectives, social media is where the supporters and opponents of specific issue discuss. This study attempts to empirically examine a recent anti-racism movement initiated by the death of George Floyd with the lens of stance prediction and aspect-based sentiment analysis (ABSA). First, this study found the stance of the tweet and users do change over the course of the protest. Furthermore, there are more users who shifted the stance compared to those who maintained the stance. Second, both supporters and opponents expressed negative sentiment more on nine extracted aspects. This indicates that there was no significant difference of sentiment among supporters and opponents and raise a caution in predicting stance based on the sentiment. The contribution of the study is two-fold. First, ABSA was explored in the context of computational social science and second, stance prediction was first attempted at scale.