Hate Speech Analysis on Gab
Our project deals with the social media website Gab, which rose to prominence recently when Pittsburgh synagogue shooter, Robert Bowers, was found to be an active user of the social media website.
But, what is Gab?
Gab is basically a Twitter clone, and just like Twitter, it allows users to read and write multimedia messages of up to 300 characters, called "gabs". The major difference in the two websites is the guidelines governing them.
Unlike Twitter, Gab has a small set of rules: No threats/terrorism, No illegal pornography, Legal pornography is allowed(but must be marked NSFW), and No doxxing
Due to Gab's very limited set of rules rules (along with poor enforcement), it has become a breeding ground for conservative, libertarian, nationalists and populist internet users.
Using Gab's API, we crawled the website and collected a total of 18562 gabs, around 10000 user profiles, 600 trending topics and 7500 trending gabs. Additionally, we gathered Robert Bower’s 123 gabs off Internet archives since his profile was deleted by Gab.
In order to detect hate speech and offensive language, we used the model described by "Automated Hate Speech Detection and the Problem of Offensive Language" by Thomas Davidson, Dana Warmsley, Michael Macy & Ingmar Weber.
Of the 18562 posts we collected on Gab, approximately 8.1% were found to be offensive and 0.9% were found to be hateful.
Some of the popular racist bigrams that we found were “n**ger n**ger”, “white supremacist” and “white men”.
Preliminary network analysis was done, and here are the major graphs/statistics which came out of our analysis.
Gab has trending “topics”, which are similar to Twitter’s “moments”. Any user can create a topic and other users can make posts under that topic. When we looked at the titles of these topics, approximately 1.2% of the topic titles were found to be offensive and around 1.3% were found to be hateful.
But when we analyzed the gabs posted under these topics, a lot more hateful and offensive posts was found.
Finally, we analyzed Robert’s Bowers gabs and ran our hate detection model on them. Approximately 3.3% of Bower’s gabs were hateful and around 6.5% of his gabs were offensive, which is approximately the same as the overall hate distribution in Gabs.
As we can clearly see, the pie chart for Robert Bowers is roughly the same as an average gab user. We can conclude that an average user on Gab spews the same amount of hate as Robert Bowers, the person responsible for Pittsburgh synagogue shooting.