Did State-sponsored Trolls Shape the 2016 US Presidential Election Discourse? Quantifying Influence on Twitter

TL;DR

Evidence is presented that the driving force of virality and influence in the network came from regular users – users who have not been classified as trolls by Twitter, and it is found that on average, troll accounts were tens of times more influential than regular users were.

Abstract

It is a widely accepted fact that state-sponsored Twitter accounts operated during the 2016 US presidential election, spreading millions of tweets with misinformation and inflammatory political content. Whether these social media campaigns of the so-called "troll" accounts were able to manipulate public opinion is still in question. Here, we quantify the influence of troll accounts on Twitter by analyzing 152.5 million tweets (by 9.9 million users) from that period. The data contain original tweets from 822 troll accounts identified as such by Twitter itself. We construct and analyse a very large interaction graph of 9.3 million nodes and 169.9 million edges using graph analysis techniques, along with a game-theoretic centrality measure. Then, we quantify the influence of all Twitter accounts on the overall information exchange as is defined by the retweet cascades. We provide a global influence ranking of all Twitter accounts and we find that one troll account appears in the top-100 and four in the top-1000. This combined with other findings presented in this paper constitute evidence that the driving force of virality and influence in the network came from regular users - users who have not been classified as trolls by Twitter. On the other hand, we find that on average, troll accounts were tens of times more influential than regular users were. Moreover, 23% and 22% of regular accounts in the top-100 and top-1000 respectively, have now been suspended by Twitter. This raises questions about their authenticity and practices during the 2016 US presidential election.

Figures (8)

• Figure 1: Toy example of retweet analysis. (a) The raw data provided by Twitter API along with the follower--graph. (b) The flow graph shows the full information flow according to Twitter functionality and the follower--graph. The edges present the path of information that appears on the users' timeline prior to their retweets. For instance, user $c$ has retweeted on date $t_{2}$. At the same time user $b$, whom user $c$ follows, has retweeted on date $t_{1}<t_{2}$. Note that a given retweet contains both the name of the user who retweeted and the name of the root user who posted the original tweet. Hence, we have an edge from the root to any retweeter because the users have retweeted the root tweet even if they did not follow the root user. (c) The time-inferred cascade tree is constructed from the flow graph by making the assumption (see Section \ref{['subsec:flow-graph-tree']}), that each retweeter has been influenced by the friend who just recently retweeted the original tweet.
• Figure 2: CCDF of the non--zero in--degree and out--degree of trolls and regular users.
• Figure 3: (a) The connected components of the undirected version of the follower--graph; (b) CCDF of the coreness for the nodes in the largest connected component.
• Figure 4: CCDF of retweet cascades in terms of unique number of retweeters and total number of retweets. The retweeters might have retweeted the same tweet more than once, hence the number of retweets is larger than the number of retweeters.
• Figure 5: Structural Virality of the retweet cascade trees.