Violent hashtag frequencies in Parler
I ran a crawler to collect hundreds of millions of posts on Parler before Amazon shut them down for hosting hate speech, conspiracy theories, threats of violence, and misinformation.
Users could only search posts on Parler (called “Parleys”) by hashtags, not by the text content within the post itself. This led users to liberally apply hashtags to any Parley that they wished to disseminate widely, providing valuable metadata to researchers like myself.
For example, here is a random Parley from January 6th in Washington D.C.
#trump had to allow #mikepence to sign his own death warrant…..
#TraitorPence #GITMO #firingsquad
#ALLorNOTHING #2theEND #wildDC #magarally #wildprotest
#saveamericarally #trumprally #dc #ArrestPence #traitorVP
#JAN6 the new #911….. #makethempay #retribution
This is the first part of a multi-part series where I analyze the data I collected from Parler. In the next part, I use this data along with FEC filings to make a case for why Georgia representative Marjorie Taylor Greene may have committed campaign finance violations.
As of March 10th, anyone can search the full dataset and download my full dataset, with all archived pictures and videos from Parler’s CDN. Thank you to all the incredible people on the Internet who helped make this possible.
Hanging politicians goes viral on Parler
The notion of murdering politicians was discussed on Parler for months before the 2020 US election, often in response to perceived injustices perpetrated by that politician. The frequency of hashtags involving hanging/execution more than doubles after the 2020 US election (blue vertical line above), more than triples on the day of the Electoral College vote (green line), and is an order of magnitude higher on the day of the Capitol riot (red line).
Parler may have allowed the idea of attacking and hanging politicians to spread virally on the platform, most likely by negligence. It is my opinion that the notion of hanging, which was gaining wider acceptance in online discourse, manifested itself in the gallows erected outside the US Capitol on January 6th. I believe the gallows are a reference to the white supremacist literature The Turner Diaries, in which a “Day of the Rope” features politicians hung en masse.
Below is a word cloud of hashtags associated with posts that explicitly mention murdering specific politicians (Nancy Pelosi, Joe Biden, AOC, and even lesser-known Democrats like Eric Swalwell).
In the graph below, hashtags are connected and placed closer together when they are frequently mentioned together. Tags are grouped into communities that are frequently mentioned together, and edges span across communities to indicate association between various topics of discussion.
Exploring the dataset
I use the Wolfram Language for analysis, which allows me to sample and search the dataset around particular topics of discussion. As I randomly perused based on hashtags in the discussion graph above, I realized how difficult moderating this kind of conversation would be. For example, consider the post below:
The author of the post is sharing a book on guerilla warfare that is available on Amazon. I can look up the post author by their user ID, and from their other posts and comments it is obvious that they are trying to incite guerilla warriors to attack the “deep state”. Amazon is wise to not provide a platform for this information to disseminate digitally, but is it simply pushing people back to the printing press?
The next viral hashtag
In the aftermath of the January 6th insurrection, there was one particular class of hashtags that appeared to be mentioned more often.
The hashtags related to civil war are frequently associated with the Boogaloo movement and the campaign of Donald Trump. It is possible we will see the real-world effects of this radicalization within the coming months.
In a world of 24-hour news cycles and the proliferation of online disinformation, hashtags are ostensibly correlated to whatever is popular on the Internet at a particular time. On Parler, most users would just “echo” the posts of others — the Parler equivalent of a retweet.
The hash tags and post IDs allow misinformation to be “contact traced” across the network. In a future post, I will share my ideas on ways to measure the contagiousness of an idea and to quantify how radicalized a user has become.
The QAnon conspiracy
There are a substantial number of Americans (including two sitting US congresswomen) who believe there is a vast global conspiracy by a cabal of Satan-worshipping pedophiles to consolidate power, and that Donald Trump is secretly trying to bring them down. One distinct feature of the QAnon conspiracy is how new layers of rationalization are continuously added through widely disseminated videos — unlike the moon landing conspiracies, QAnon’s purported conspiracy is ongoing, and continuously adapting to current events. As the theory evolves, it provides opportunities for anyone on the Internet to insert more dangerous ideas into it.
This is a popular QAnon post taken from a random sample of the dataset. It has been viewed over 181,000 times, reposted over a thousand times, upvoted over a thousand times, and has 319 comments attached to it.
You can now read part 2 in which the data indicates that a political campaign and a super PAC shoveled over a quarter million dollars into Parler in the span of one month after the election.
The Internet Archive is putting together a far more comprehensive dataset of posts, images, comments, users, and media from Parler which I recommend to data scientists looking to take up this project. My future plans for this project are:
- Perform NLP and sentiment analysis on the comments that users leave on posts, and identify which hashtags are associated with which sentiments
- Analysis of disinformation sites, the rate at which they propagate across the Parler social network, and the hashtags most frequently associated with each one
- Quantifying the radicalization of a user by showing a statistically-significant trend toward more violent ideas and hashtags
- Clean up the data and make it publicly accessible to researchers
You can contact me about this post and to request access to the dataset by messaging me on Twitter or emailing email@example.com. I am trying to stay anonymous so these stories aren’t about me — they should be about the data. If you read anything else I write and find a partisan lean to it, please call me out on it, as it defeats the whole point of writing these if I’m just speaking to an echo chamber of my own.
Please don’t make the problem worse
I do not think the average Parler user deserves to be victimized, called a fascist, or have their privacy taken from them. I will not be taking part in anything like that.
The people who committed crimes should be treated as criminals. However, there were some pretty sociopathic people on Parler who need compassion and possibly mental help, not further provocation. I believe we need to study how technology enabled a group of people to radicalize each other and reinforced an echo chamber where conspiracies and racism could flourish.
Disclaimer: since Parler has never enforced their Terms of Service with regards to violent speech, I slept fine at night ignoring their ToS with regards to automated scraping of the site.