Automatic Fake News Detection on Social Networks using Multimodal Approach of BERT and ResNet110

With the rapid development of online content-sharing and social media platforms has exploded for spreading fake news which create people trusting in unfavorable stories. This scenario allows to determine public opinion and produce chaos and panic through population. Therefore, detecting fake news has one of most significant topics which aims to identify a certain news is real or fake. The solution of fake news detection is split into three various classes such as content, social context and knowledge-based techniques. In this paper, a multimodal approach is proposed for detecting fake news for text and image, as well as classifying the news is real or fake. The Bidirectional Encoder Representations from Transformers (BERT) is used to extract text features and ResNet110 is used for image feature extraction. This model is estimated on Buzzfeed and ISOT dataset and attains better results using accuracy, precision, recall, specificity, and F1-score. The obtained result shows that the BERT+ResNet110 model achieves better accuracy of 98.51% on Buzzfeed dataset and 99.68% on ISOT dataset which ensures accurate detection compared to other existing methods like Random Forest (RF), Multi Support Vector Machine (MSVM) and Long Short-term Memory-Levy Flight (LSTM-LF).