Can Deep Learning Detect Fake News Better when Adding Context Features?

Fake news is increasing on social media and has huge negative impacts on society. To detect fake news, different approaches and algorithms have been studied. Recently, the use of deep learning algorithms has performed very well for fake news detection. Many researchers found the importance to add the context features to have better and a rapid detection. The current paper presents two algorithms, BERT and LSTM, for the classification of fake news, by using text and context features. Moreover, we add XAI solution that determines which context features have contributed the most to the classification of fake news. We have trained our model with an extraction of 5,000 news from the COVID-19 dataset. Our result shows that BERT performs better by 2.37% than LSTM by analyzing only the content and performs better by 2.24% by analyzing both, the content and the context. Finally, SHAP is used to explain fake news by highlighting the most relevant attributes that have helped to the classification of news.