Using Machine Learning Algorithms to Detect Fake News

Abstract

Fake news has been a growing threat in the modern world. A major reason why fake news is so dangerous and effective is due to the difficulties of distinguishing it from correct news, if there was a way to detect fake news accurately, its negative impact could be significantly minimized. Previous studies have already found that fake news differentiated itself substantially from real news in terms of words used and the structure of the texts, implying the possibility of differentiation. One possible method of detecting fake news is Machine Learning. Utilizing artificial intelligence to detect patterns within the text of fake and real news articles. In this paper, we test the capability of the Machine Learning Algorithms in detecting fake news using four different types of models, SVM, Multinomial NB, Gradient Boosting, and Gradient Boosting with LDA. We find that all four models had a high success rate of over 90%, with the LDA+Gradient Boosting model performing the best, and Multinomial NB being the least successful. We also attempt to determine the topics that fake news tends to cover and found that fake news is often about politics. While the model has proven to be successful, we recommend that future testing be done on other datasets with greater variety in news sources.