Bot or Not? Algorithm-driven Public Opinion During the Global Coronavirus Crisis


The concern over manipulation of vox populi in the digital and algorithmic communication environment has motivated burgeoning research on understanding how bots grow into an emerging power in shaping public opinion and blurring civic conversation. While the influence of bots has received considerable attention in the context of political communication, very few studies examine bots in the public health contexts, especially in epidemics where the urgency and accuracy of information are at high stakes and where the risk travels beyond national borders and affects global communities. In this study, we examine the scope and patterns of influence of Twitter bots during the outbreak of coronavirus in 2019-2020.

We first ask: What is the scope of bot-generated information in the Twitter discourse on coronavirus outbreak (RQ1)? Further, we ask: what are the distinctive behavioral patterns (RQ2), topical focuses (RQ3), and network features (RQ4) of Twitter bots, compared to human accounts? How do these behavioral patterns, topical focuses and network features change overtime (RQ5)? Finally, we ask: How much does the salience of topics discussed by Twitter bots influence salience of topics discussed by (a) human accounts and (b) mainstream media (RQ6)?

This analysis focused on English-only content generated by human and bot-like accounts center around the topic of coronavirus outbreak on Twitter. Tweets published from December 1st to the end of February are collected.

Two cutting-edge bot detection systems, Botometer and tweetbotornot, are applied to differentiate bot-like accounts from human accounts based on our tweet dataset. For those accounts with an assigned bot score higher than a threshold value will be labeled as a bot account. Topic modeling techniques are used to discover the hidden topics embedded in tweets generated by humans and bots. Network analyses are employed to discover features of networks Twitter bots are embedded in. Time series analysis are used to examine the longitudinal patterns of Twitter bots’ behavioral patterns and topical focuses, as well as the relationship between topics discussed by bots, human accounts and mainstream media.

We initially found that bots have distinct behavior patterns in tweeting with coronavirus-related hashtags, for instance, generating a large proportion of commercial ads and pornographic contents but a less public health-related discussion. Besides, some political-oriented hashtags (e.g., #Chinalies, # Chinazivirus) were partially driven by bots. This phenomenon shows an implication of a potential spill-over effect on the political conversation of which goes beyond the original focus on public health issues. Further findings and limitations were also included at the end of our study.