In this month’s research spotlight, COSMOS highlights recent research that investigates the temporal distribution of topics within digital echo chambers. Specifically, the study authored by Dr. Esther Mead, Hillary Woodworth, and Dr. Nitin Agarwal titled, “Text Mining Domestic Extremism Topics on Multiple Social Media Platforms” looked at “Balkanized” (i.e., divided, mutually hostile groups) digital spaces utilized by certain social media actors on Parler, Twitter, and YouTube. The study was published at the 6th International Conference on Data Mining and Knowledge Discovery (DMKD 2024), which took place virtually from June 7 to 9, and at the 5th International Conference on Computational Linguistics and Natural Language Processing (CLNLP 2024), which was held in Yinchuan, China from July 19 to 21.

The “Balkanization” of virtual communities into partisan echo chambers is highly effective in curating polarized perceptions of real-world events. The authors draw upon Uses and Gratifications Theory (UGT), which recognizes that users select media that meets their psychological or social needs, such as desire for information, emotional connection, and status. 

“Considering that media channels continue to become more isolated and fragmented based on polarized norms and behaviors, it is imperative that researchers are able to take a close-up look at distinct social worlds that exist in these ‘Balkanized’ digital spaces that pit social capital and connectivity against societal expectations and norms,” they explain. For this reason, the authors chose not only to use Latent Dirichlet Allocation (LDA) topic modeling for finding themes but chose a multi-platform or “transmedial” approach that would find common themes or differences across the multiple platforms of Parler, Twitter, and YouTube for the time period surrounding the insurrection. In addition to topic modeling, toxicity of posts was analyzed. 

Notably, these different platforms had major differences in posted topics and toxicity, possibly showcasing that different platforms had differing levels of Balkanization of partisanship. 

They concluded that extremism on social media is characterized by specific narratives and a high level of toxicity. Influential accounts play a significant role in disseminating extremist views and increasing the toxicity of online discourse. These findings emphasize the importance of understanding the dynamics of polarized discourse on social media platforms and the need for further research to develop strategies to counteract extremist narratives. Moreover, the study’s use of text mining techniques and LDA topic modeling offered a comprehensive understanding of the dominant narratives and toxicity levels on Parler, Twitter, and YouTube. 

Dr. Agarwal said, “This research highlights the role of social media influencers in shaping online discourse with broad implications for our society. With 2024 being the year of elections worldwide, such research shows the power of social influence that propagates in online and polarized communities.”