Towards better representation learning using hybrid deep learning model for fake news detection

Detection of Fake news articles over the internet is a difficult task due to huge amount of content being proliferated. Fake news proliferation is a major issue as it has socio-political impacts and it may change the opinion of the people. The easy dissemination of information through social media has added to exponential growth of fake news. Thus, it is challenging task to detect the fake news on the internet. In the literature, fake news detection techniques have been developed using machine learning approaches. Usually, fake news consists of sequential data. Recently, different variants of the Recurrent Neural Networks have been used for fake news detection due to better handling of sequential data and preserving better context information. Due to diversity in fake news data there is still need to develop the fake news detection techniques with better performance. In this work, we have developed a hybrid fake news detection model which aims at better representation learning to enhance the fake news detection performance. The proposed hybrid model has been developed using N-gram with TF-IDF to extract the content-based features then sequential features have been extracted using deep learning model [LSTM or bidirectional Encoder representation from transformers (BERT)]. The performance of the proposed approach has been evaluated using two publicly available datasets. It is observed from results that the proposed approach performs better the fake news detection approaches developed in the literature. The proposed approach has given the accuracies of 96.8% and 94% for the WELFAKE and KaggleFakeNews datasets, respectively.