/fakenews

A collection of labeled fake news and real news (top credible news sources from https://webhose.io/) was used to train a Naive Bayes model to predict probabilities of fake news based on article text. The model predicted fake or real with 85% accuracy using ~1,100 articles in our testing set and a training set of ~4,000 articles. We cannot guarantee the correctness of our labels, given the subjectivity of the terms "fake" and "real". Therefore this tool is not a perfect judge of all articles, but can be used as a gentle guide.

Primary LanguageJupyter Notebook

Stargazers