How to detect cross-platform misinformation campaigns (part 1)

Author: Zhouhan Chen
Date: July, 2022
Information Tracer Home Page

(Top: visualization of an information campaign sharing Breitbart news URL. Node with a URL icon represents a URL; node with a brand logo represents an account from that social network platform. Two nodes are connected if an account has shared that URL before. Here we see accounts from multiple platforms repeated sharing URLs from Breitbart, a far-right website with questionable source.)

Our Motivation

• Since 2020, Information Tracer has been collecting and tracing more than 100,000 news URLs over 6 different social media platforms – Twitter, Facebook, YouTube, Reddit, Telegram and Gab.
• In this blog, we will show you how to transform data into action by discovering and visualizing cross-platform information campaigns.

Potential Impact

• As our previous research demonstrates, there is a proliferation of fake news since the 2020 COVID-19 Pandemic. Fake news is often posted on one platform, then reshared on others. In short, focusing on one platform is not enough.
• Understanding cross-platform misinformation campaign allows us to detect fake news early and stop fake news from spreading too wide and too far.

Method: Automatic Campaign Discovery Using Locality Sensitive Hashing (LSH)

We now give a brief overview of our discovery method. If you want to see our findings, feel free to skip this section.

1. To find cross-platform information operations, we assume that there are multiple accounts from different platforms sharing similar content. Now our task becomes – given the data we collected (URLs, accounts sharing each URL), how to find groups of accounts that share similar URLs?

2. To measure account similarity, we use Locality Sensitive Hashing (LSH). LSH is an efficient algorithm to approximate Jaccard similarity (similarity between two sets).

   To give you a concrete example, if URL true[.]com is shared by accounts Alice, Bob, Chris, and URL false[.]com is shared by Bob, Chris, David, then Jaccard(true[.]com, false[.]com) = |{Alice, Bob}| / |{Alice, Bob, Chris, David}| = 0.5

3. Next we build a graph. Each node is an account. Two nodes are connected if their Jaccard similarity is higher than a threashold.

4. We extract connected components from the graph. Each connected component is a group of URLs shared by accounts from different platforms. We do a reverse search to find the most common accounts on each platform that share those URLs. This group of account forms the basis of a campaign.

5. To identify misinformation campaigns, we use Media Bias Fact Check and other sources to assess the quality of each URL domain /news organization. If a news organization has repeatedly shared propoganda, conspiracy or false information, we label the group of accounts sharing their URLs a misiformation campaign.

Our findings

We identified a large number of cross-platform campaigns sharing fake or misleading news. In this blog we will highlight four news organizations. We will share more findings in future blogs.

[1] Breitbart

Breitbart is labeled questionable source due to its extreme right content, propaganda, conspiracy, and failed fact checks practice. According to our findings, Breitbart articles are co-shared on two official accounts – breitbart on Facebook and breitbart on Telegram. On Reddit, Breitbart is promoted by accounts such as republican and conservative. On Twitter and Gab its shared by dfbharvard, an NRA-ProLife-Ultra-MAGA-Conservative. To see a a recent URL spreading pattern on Information Tracer please visit here

[2] The Federalist

The Federalist is another questionable source posting conspiracy theories, propaganda, and has failed fact checks. The website has three major spreaders – the federalist on Facebook, fdrlst on Twitter, and the federalist on Reddit. On Gab several accounts including f16viper01, williammbarrett, ken19512 also spread its URLs. To see a a recent URL spreading pattern on Information Tracer please visit here

[3] Todd Starnes

Similar to the Federalist, Todd Starnes is another far-right news media with poor sourcing and lack of transparency. The website promotes conspiracy and propaganda without adequate fact checks. Our data collection reveals that the website is shared on at least 4 social media platforms – Facebook (by todd starnes and the mighty 990), Gab (by toddstarnes), Reddit (by todd starnes) and Twitter (by toddstarnes). To see a a recent URL spreading pattern on Information Tracer please visit here

[4] The Western Journal

If previous examples are borderline suspicious, the last example is much more egregious. Rated as questionable, The Western Journal has numerous failed fact checks. Unlike previous fake news websites with only a few related accounts, The Western Journal employs an army of accounts.

Based on our data collection, we identified at least 12 Facebook accounts repeatedly sharing URLs from The Western Journal. On Reddit we identified 14 highly similar accounts. Those accoutns use variations of The Western Journal as account names – such as western journal opinion, the tea party community by wj or patriot update by wj. Below is a list of accounts we identified, as well as their platforms. To see a a recent URL spreading pattern on Information Tracer please visit here.

What’s our next step?

1. Intelligence sharing. We hope our findings can promote social media companies to launch in-depth investigations into those accounts and behaviors. Social media companies can use their internal data (such as IP address, register email, or other strong signals) to determine if there is actual coordinated manipluation. Based on our current data collection, we are not able to attribute the owner of each account, or if an account is controlled by a human or a bot (an automated script).
2. Improving detection capability. Finding coordinated manipulation is highly non-trivial. What we showed is probably just a tip of an iceburg of all suspicious cross-platform activities. We will continue refining our detection algorithms and publish our findings whenever appropriate. If you are an organization or individual who wants to collaborate, please feel free to reach out to us at zhouhan.chen@nyu.edu.
3. This article is part one of a multi-series blog on detecting cross-platform misinformation campaigns. To learn more please check out part two.