Fake News Detection Using Deep Learning and Transformer-Based Model

Fake news has a tremendous impact especially in certain fields like politics and economy in our society. The rise of social media usage to an extent is in favour of fake/false news promotion. A strong solution in this area would have numerous advantages for the society as a whole. Various deep learning techniques have been suggested by researchers to address this issue. But the majority of them don't identify false news with the requisite precision. In this paper, we propose a fake news identification model with state of art accuracy. Along with that, we compare different deep learning models in combination with different word embedding techniques for the task of fake news detection. Two types of word embeddings, Glove and Word2Vec were used for converting text to a vector of numbers. Along with the word embedding techniques we used seven different combinations of deep learning models: RNN, LSTM, Bidirectional LSTM, GRU, Bi-directional GRU, CNN-LSTM, and CNN-Bi-directional LSTM. The proposed work also implemented FastText and BERT models for embedding and classification of false news. We discovered that the proposed BERT classification model performed most effectively for detecting fake news on the experimented dataset. The proposed BERT model outperformed various baseline models on the benchmark dataset, attaining a test data accuracy of 99.20%, establishing its state-of-the-art performance. For the evaluation of models, we used accuracy, precision, recall, and F1-score metrics on the test data.