A Performance Comparison of Fake News Detection Approaches

Abstract

Fake news occurs when a news article is intentionally released and subsequently shown to be false. This can negatively impact on an individual’s or indeed society’s comprehension of topics and shape public opinion on major events, e.g. elections. With the ubiquity of social media and fake news websites, people are increasingly exposed to fake news. In this article, we explore and compare the performance of fake news detection approaches using both machine and deep learning methods. Specifically we explore Logistic Regression, Support Vector Machines (SVM), Recurrent Neural Networks (RNN), Long Short Term Memory (LSTM) and Bidirectional Encoder Representations from Transformers (BERT) and compare the results with human crowds ability to distinguish real vs fake news. We utilise over 8.5m records from FakeNewsCorpus and 23k records from FakeNewsNet. The results shown that BERT achieved the best overall accuracy at 82.5%. Groups of individual assessors achieved an accuracy of 79%, whilst individuals varied significantly between 60-80% in their ability to distinguish real vs fake news.