Disinformation and Social Bot Operations in the Run Up to the 2017 French Presidential Election
Article Link: https://arxiv.org/pdf/1707.00086.pdf
This article reports on a pattern of disinformation spread through bots on Twitter in the French 2017 election. By trying to determine which accounts were bots, patterns could be found in bot identities and behavior, leading to evidence suggesting that bots for spreading disinformation were available in a black market.
In order to separate humans from bots, they compared accounts’ behavior to that expected by a human. This is a more complex version of the process we used in finding fake/hacked social media accounts in Chapter 16. We used Bayes’ Rule given a predicted probability of legitimate interest in a site given a click in order to determine the probability that the person had a legitimate interest. The analysis used in this paper involves a much more complex and thorough model of human behavior, as well as machine learning in order to better account for the unaccounted variables in their model. While this paper used a much more complex method, it most likely used Bayes’ Rule and probability similarly to how we used those in class.
This paper also considered the effect of cascades in disinformation. In order for a bot to spread disinformation effectively, it must overcome the correct sources of information. In other words, it must create an information cascade in which everyone believes that the probability that the incorrect fact is true is higher than the probability that the correct fact is true. In the information cascades that studied in class, if this happened, then the incorrect fact would be spread to everyone. However, in this real life situation, people have different amounts of information, signals are not just introduced one at a time to each person, and not all signals have the same effect (for example, reputable news sources would have a greater effect than a high school student’s Tweet). Nonetheless, similar to what we did in class, this paper traced information cascades to guess what each person’s signals were, and how the bots used information cascades to their advantage.