[Air-L] Trust Defender

[Shulman, Stu]

Trust Defender is a collection of open source scripts used to train and run
a classifier to classify Twitter users as potential good or bad actors. The
classifier models are based on Twitter user information and focus heavily
on the user bio description to make determinations.

https://github.com/texifter/trust-defender

There are two primary classifier types: an n-gram naive Bayes classifier
model, which uses 2,3,4, and 5-gram models combined and results a
probability score for two classes: bot or good, a neural network that takes
as input: p(bot) (from  the n-gram classifier) p(good) (from the n-gram
classifier), number of days the Twitter account has been active, average
number of statuses per day, average number of followers per day, average
number of per day, the description length in number of individual terms,
the number of “lists” in the description (e.g. god, country, president
would be considered a “list”), the number of hashtags used in the
description, the number of URLs found in the description.

Not every seemingly odd account is a "bad actor" or a bot, however, you can
find some pretty freaky stuff using this model. Happy hunting. Elon is
waiting for your findings.

-- 
Dr. Stuart W. Shulman
Founder and CEO, Texifter
Editor Emeritus, *Journal of Information Technology & Politics*