When blocking hate speech, we should look at the source, not the words.
Social media has played an outsize role in the 2016 election, with Donald Trump's supporters using formerly respectable internet memes like Pepe the Frog to spread a racist message.
Sometimes, these memes come from dark money-backed content factories funded by tech moguls (looking at you, Palmer Luckey). But more often than not, like Pepe, they come from 4chan.
To the uninitiated, 4chan can best be described as a bulging cyst on the ass of the internet. It's often recognized as a bastion of the alt-right, the main faction boosting Trump's candidacy online.
Trolls from 4chan sometimes orchestrate ugly and racist posting campaigns on the wider web and attract media attention as a result, leaving experts and tech companies wondering how to stem the online tide of politically-motivated hate. Instead of trying to filter keywords used by racists, which are always changing, researchers want to block the source: 4chan.
A team of researchers funded by the European Union's ENCASE project, which seeks to improve social networks, have completed a large study of user activity on 4chan's "Politically Incorrect" board, /pol/, and invented a mechanism to detect if trolls from that site are invading other social media sites to spread hate as part of a targeted campaign.
"One way to look at countermeasures to hate speech, for Google or for Twitter, is to look at activities on 4chan. Because we do find evidence that some of the hate speech is coming from 4chan," Emiliano De Cristofaro, a computer scientist at University College London and one of the authors of the study, told me in an interview.
One of the goals of the EU's ENCASE program is to invent browser add-ons for users and server-side software for companies in order to enhance security and privacy on social networks.
"It's much more effective to look at groups and where the content's coming from, instead of keywords"
In the study, which hasn't been peer-reviewed, the researchers scraped 4chan between June and September, amassing a corpus of eight million posts and over 200,000 threads. Analyzing these, the team was able to correlate upticks in hate speech in the comments of certain YouTube videos to the time the link was posted in a thread on /pol/. The team also looked at Twitter's firehose of tweets to see if a particular hashtag was being co-opted by 4chan trolls.
De Cristofaro and his colleagues used this approach to analyze the effect of "Operation Google," a campaign launched on /pol/ that asked participants to replace common racial slurs with the names of Google products and other code words. The goal was to get around Google's algorithmic hate speech filtering. According to the researchers' study, the trolls' campaign managed to infiltrate YouTube, but was otherwise largely confined to 4chan itself.
"When filtering out hate speech automatically, it's much more effective to look at groups and where the content's coming from, instead of keywords," said De Cristofaro on Operation Google. "The keyword approach can be evaded by switching the terms or using code-words."
There's something to be said for just ignoring 4chan, /pol/, the alt-right, and trolls altogether. While these groups may have a loud voice online, their champion—Trump—is slipping in the polls regardless as gaffe after gaffe continue to convince much of the public that, amazingly, he's exactly as bad as he seems upon first inspection. As troll scholar Whitney Phillips noted in a recent essay for Motherboard, media attention (admittedly, just like this article) is the lifeblood of the movement.
But their hate does spread beyond the online world where it proliferates. It's worth mentioning, for instance, that the most popular sites shared on /pol/ besides YouTube and Wikipedia, according to De Cristofaro and his colleagues' study, were the less-than-reputable right wing "news" outlet Breitbart, and Donald Trump's own website. A recent Washington Post profile on a mentally ill Trump supporter illustrated how the half-truths and toxic attitudes that ooze from such sites often buoy pro-Trump sentiment.
"The alt-right is using 4chan as a source for information," DeCristofaro said. "Whether you like it or not, 4chan is very popular and has a lot of influence."
Get six of our favorite Motherboard stories every day by signing up for our newsletter.