This is a Twilio/Flask application built to estimate the probability that a news article is "fake". It also utilizes Bing's Spell Checker API to get number of typos and add it as a feature! The algorithm is a Random Forest Classifier based on the sklearn open source code!!
...A product of the Global AI Hackathon NYC 2017...
What Austd Really Is
We don't want to call news "fake" or "real". This tool is really just a way for you to check the esimated reliability of your news article. It gives you a guess as to whether or not you can rely on it along with a percentage of confidence. At the end of the day, you have to decide whether or not to let your confirmation bias take you over.
Our model is based on a few key features that are derived from the text and the title. It is not based on the text directly. This is on purpose, because then we fear that the topics within the text itself will have more weight than the other features in terms of the wordsmithery of the text. This proved to be quite accurate in the Random Forest model (see the notebook). Please contact us for more advice on future directions, however. Our true knowledge exists in that we know that we know nothing!
As an added feature, we used OpenSources, a professionally curated list, to tag sources that have been previously identified as fake, false, extremely bias, generally satiric, and more. Detailed methods for tagging are on: http://www.opensources.co/
How To Use It
Go to our page! (this section will be updated once we have a link.)
Or text it! (this will be updated too.)
Enter a url to a news link!
You will get your answer back! plus a cute gif :)
Improve/enlarge training dataset. it is currently biased by the fact that the "real" news is all British.
Develop a Human-in-the-loop role so that verified contributors can upvote or downvote articles.
Include more features?
Subjectivity Lexicon from the CS department at Pitt.
"Fake" News from Kaggle.
VADER algorithm used to measure neutrality in titles.
"Real" news from Multimedia Knowledge and Social Media Analytics Laboratory.