Project for the Practical Neural Networks (DS 4440) class at Northeastern University, Khoury College, during Fall 2020
The popularization of social media and the democratization of the ability to cheaply broadcast information, deceptive content, including campaigns for intentional misinformation and disinformation, have become problematic. To keep the web a safe place, we require a scalable method to limit the spread of malicious content, and "fake news" in particular. The natural rapid change and rapid emergence of new topics around social media and news articles causes models to quickly become obsolete. This paper explores a Fake News Detection using a method base on BERT and Adversarial Neural Networks for normalization against learning features to speci�c news topics/events. We compare this to a non-adversarial model and analyze performance results and bias on key topics.
Please request access to data from author ([email protected]).