Examining the Influences of Twitter Bots during the 2022 U.S. Midterm Election

• In Event: The Big Lie on Social Media

Abstract

Social media has been a battleground among partisans across all contexts and topics. One of the emerging entrants that are concerned to make the online environment more polarized is social bots. Many studies showed that bots are politicized along the partisan line and strategically enhance the visibility of one party. In the U.S. context, bots were found to promote messages from the political right. Regarding the potential threats of bots in polarizing online discussions, existing studies mainly focus on the descriptive studies of comparing the discourses between humans and bots or merely analyzing the network structure with bot activities. However, little is known about whether bots are a real threat to the online environment through investigating the effectiveness of bot activities. Thus, we are going to ask whether bots are effective enough to involve more conversations from a human-bot interaction perspective. Specifically, other than examining what topics the bots promoted, we will calculate the engagement capacity of bots by looking into the social network structure and the retweeting cascading with other users. This paper will provide insights into the influentialness and effectiveness of Twitter bots during the 2022 U.S. midterm election through a social network analysis approach.
The position of bots in the network is one factor to estimate the importance of bots. Typically, the more central the position, the more influential it is. Researchers found that bots mostly situate in the peripheral position of the network and tend to target the human “hub” of a network. When looking into the political differences, it was found that compared to liberal bots, conservative bots tend to sit in the central position and are well-embedded in the network. Considering Twitter bots are mainly amplifiers, we focus on the position of bot accounts by constructing a retweet network (RQ1).
Previous studies found that bots may spread misinformation/disinformation by promoting content from partisan outlets or low-credibility sources. On the other hand, sometimes, bots are just news aggregators who retweet news from mainstream media. However, most of the studies didn’t further look into the bots’ targets from the user level. In this study, we interrogate the primary target groups of bots by analyzing the user descriptions of retweeted accounts (RQ2).
In recent years, researchers attempt to demystify whether bots take control of online political discussions or not. One line of study is to compare the discourses promoted by bots and humans in specific contexts (e.g. Covid-19, elections, and climate change). Along this line, we first ask: Do bots promote the same or different discourses than humans during the 2022 midterm election? (RQ3) Apart from discourses, the more critical question to ask is whether bots are influential enough to engage other users, or whether we over-concern the effectiveness of bot activities. We will use the method proposed by Nikolaev et al. (2016) to analyze the retweet cascading chain of bots accounts and measure the engagement capacity of Twitter bots. In that way, we are expecting to measure whether bots are capable to involve more users through retweeting activities (RQ4). We believe our method will be valid and pertinent to present the human-bot interaction relationship and the effectiveness of bots.
Twitter data were collected by Twitter Academic API v2, using search terms “midterm” or “2022midterm” from 8/1/2022 to 10/20/2022. Before deduplication, it ended up with more than 3 million tweets, and more than 1 million unique users. We will employ Botometer v4 API to check the botnets of accounts. A recent study found that v4 has a very high accuracy in detecting bots compared to previous versions. Next, we will create a retweet network to analyze bot position and use community detection to identify the retweeted target groups of bots based on the originator’s user descriptions. Then, automated text analysis will be conducted to analyze the discourses. Lastly, we will construct the retweet cascading chain that is inferred from time spots and the number of followers and estimate the engagement capacity of bot accounts.

References
Nikolaev, A., Gore, S., & Govindaraju, V. (2016). Engagement Capacity and Engaging Team Formation for Reach Maximization of Online Social Media Platforms. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 225–234. https://doi.org/10.1145/2939672.2939681

Authors

• Rui Wang, University at Buffalo

• Alexander Nikolaev, University at Buffalo

• Alexander Semenov, University of Florida

• Yotam Ophir, University at Buffalo