Hyperpartisan News and Articles Detection Using BERT and ELMo

Abstract

Fake news and articles are misleading the readers. This leads to the increasing studies of fake news article detection over the decades. Hyperpartisan news is news riddled with twisted and untruth and extremely one-sided. This news can spread more successfully than others. Besides that, hyperpartisan news can mimic the form of regular news articles. This study aims to identify and classify the hyperpartisan news with BERT and ELMo. Two distinct models, BERT and ELMo, were created to classify hyperpartisan news from two datasets, namely by-article and by-publisher. Few other models with different settings and training designed to test and optimise the performance of both models. The results of the optimised BERT and ELMo models can achieve 68.4% and 60.8%, respectively.