Ant-BERT Based Optimized Fake News Detection

The widespread and rapid spreading of false information on social media become a prominent concern in outside society, driven by the ease of disseminating and consuming misleading information. In recent research, many models came up with unique approaches to classify input as fake or true and those techniques are unidirectional and not context-based models. In this work, a new novel methodology using BERT and Ant Colony Optimization (Ant-BERT) was proposed for fake news detection. As a sentence encoder, BERT could able to precisely extract a phrase's context representation by extracting word embeddings. Further, Ant Colony optimization is employed to reduce the number of embeddings. Finally, the output vector is multiplied with the reduced embedding matrix and then a dense layer with the Softmax function is employed at the top to compute the likelihood of each vector. Classification results of Ant-BERT clearly evident superior performance of 95.5% accuracy on fake news detection dataset when compared to other Context-based deep learning models.