Using natural language to predict bias and factuality in media with a study on rationalization

Abstract

Fake news is a widespread problem due to the ease of information spread online, and its ability to deceive large populations with intentionally false information. The damage it causes is exacerbated by its political links and loaded language, which make it polarizing in nature, and preys on peoples' psychological biases to make it more believable and viral. In order to dampen the influence of fake news, organizations have begun to manually tag, or develop systems to automatically tag, false and biased information. However, manual efforts struggle to keep up with the rate at which content is published, and automated methods provide very little explanation to convince people of their validity. In an effort to address these issues, we present a system to classify media sources' political bias and factuality levels by analyzing the language that gives fake news its contagious and damaging power. Additionally, we survey potential approaches for increasing the transparency of black-box fake news detection methods.