Detecting Fake News Sources on Twitter Using Deep Neural Network

Abstract

Social media provides a rapid, simple, and accessible platform for people to communicate and share news through the Internet. However, the information published on this platform is not always trustworthy. As a result, malicious actors often use social media to disseminate fake news or mislead news readers, such as with personal or political attacks that could spark protests or riots. In this paper, we propose a learning technique for detecting fake news sources (i.e., fake users) on the Twitter platform. Three main types of features—tweet content, published time, and social graph—have been defined and extracted from Twitter to create a deep neural network (DNN) as a predictive model. We conducted experiments on PolitiFact, a standard FakeNewsNet dataset. The results show that the proposed approach outperforms traditional baselines with 98.7% accuracy.