Figure 7

Impact of humans and bots. (a) Distribution of bot scores for source tweets and retweets. The two groups had significantly different bot scores (Mann-Whitney U test: \(U = 2 \cdot 10^{9}\); \(p < 0.001\), and Mood’s median test: \(\chi = 666.84\); \(p < 0.001\) with median bot score of source tweeters = 0.22 and median bot score of retweeters = 0.27), implying that retweeters were more likely to be bots. (b) Joint distribution of bot scores of authors of source tweet-retweet pairs (heatmap). The top subplot shows the distribution of bot scores for retweeters. The left subplot shows the distribution of bot scores for accounts that were retweeted by accounts classified as humans (using a threshold of 0.5). We find that most source tweets were posted by humans. They were also active retweeters, but so were bots. Different from the spread of low-credibility content [20], we find that a significant proportion of retweeters were bots and that they tended to retweet humans rather than other bots.