Under our AI for Good initiative, KUNGFU.AI has analyzed tweets related to the Coronavirus/COVID-19 pandemic in order to find possible patterns of disinformation and unusual propagation of content on Twitter. In separate time periods starting in early February 2020, we have analyzed more than 19.95M tweets that mention Coronavirus and/or COVID-19 using the free Twitter public API. We used information entropy measures calculated from both the text content and the timing of the tweets to identify accounts displaying significantly more automation than normal Twitter users. To date, we have identified 3,378 such automated accounts. We then applied network graph analysis techniques to determine which automated accounts were most successful in getting their content shared or retweeted. Consistent with our prior research on automated Twitter accounts in the political realm, we find that cyborgs, actual humans who use special software to control or automate their posts, are generally much more successful than fully automated bots. For more information on our research methods, see the “Identifying viral bots and cyborgs in social media” O’Reilly Data blog post by KUNGFU.AI’s chief scientist, Dr. Steve Kramer.
Our key finding is that these automated accounts can be separated into the following groups:
As has been reported previously, there are numerous Twitter accounts that claim that the Coronavirus is fake news or a hoax, and social media companies like Twitter, Facebook, and YouTube are attempting to detect and suppress those types of disinformation campaigns. In the Twitter data sets we analyzed, many of the viral tweets that include the phrase “Fake News” are actually admonitions from news and health organizations warning about the proliferation of disinformation regarding the Coronavirus.
We did, however, find examples of dense networks of automated accounts that tweeted incorrectly that COVID-19 is “just the flu” instead of the risky novel virus that it is. The figure below highlights such a network centered around the Twitter account @IsChinar, which has been suspended by Twitter since we collected these tweets in early February. Note the numerous deceptive tweets implying that COVID-19 is “just the flu” centered around that malicious account. This interactive network graph visualization is freely available online on polinode.com.
We performed a historical query for the automated accounts we identified and then expanded the shortened URLs in order to find out which web domains (external to twitter.com) were referenced by the bots/cyborgs. The pie chart below shows that rt.com, youtube.com, and sputniknews.com were the three domains with the greatest number of URLs linked to by the automated accounts we studied. In our experience, it is not unusual for YouTube URLs to abound in Twitter data sets, but we were surprised by the large number of URLs for rt.com and sputniknews.com, which are influential Russian news websites. Because it can be extremely difficult to ascertain which persons or organizations manage any specific Twitter account, we are certainly not claiming that these accounts are controlled by state actors. However, because the Coronavirus pandemic is a global issue, the predominance of these Russian websites in this bot/cyborg analysis is notable.
The network graph below summarizes the activity of the bots/cyborgs that linked to these Russian websites. Note the appearance of separate clusters based on the topics the users tweet about and the other users they mention or retweet.
Two specific examples shown below give evidence that at least some of these accounts are supportive of Donald Trump and the Republican party in general:
In most of the time periods during which we collected and analyzed tweets, we found the most successful automated accounts are those that retweet Coronavirus news from mainstream news websites. As examples, the summary tweets from the top-ranked automated accounts from February 2-15, 2020 and from March 8-11, 2020, respectively, are shown below. In many cases, one would rightly be concerned about the activities of automated Twitter accounts in terms of the risks of spreading disinformation. However, we believe that this category of automated accounts are primarily a beneficial force that is acting to spread mainstream news as the stories evolve.
Especially during a second time period from February 22-25, 2020, we found that among the most viral automated accounts, there was a separate cluster, or subnetwork (shown on the right in the figure below), centered around the @SpectatorIndex Twitter account. This account posts its own content about financial data, recently primarily about Coronavirus and its effects on the financial markets. Its linked website encourages readers to sign up for its paid subscription, which is, of course, a standard business practice. What is unusual is the cluster of automated accounts that retweeted @SpectatorIndex’s tweets, which successfully drove online attention to that commercial account.
GIven the urgent nature and challenges associated with the COVID-19 pandemic, we wanted to share our preliminary research findings. We have made subsets of our data and results available on data.world, to the extent allowable by Twitter’s Terms of Service. We are also making an open call to collaborate with researchers in universities, companies, and organizations committed to raising awareness about the risks of disinformation in this critical time. Please email firstname.lastname@example.org with “COVID-19 Research Collaboration” as the email subject line if you are interested.