A hybrid deep learning approach for binary classification of fake news from social medias' multimodal data

Abstract

In this digital age, social media sites are very important for spreading news. However, they are also great places for fake news to grow, which spreads false information widely and causes problems in society. This thesis tries to solve the problem of classifying fake news by combining different types of data and using BERT and DistilBERT to process text and ResNet34 and ResNet50 to process images. A lot of experiments were done and the obtained results showed that the BERT + ResNet50 model worked best, getting a high accuracy rate of 94%. Textual and visual data are captured and combined very well in this way, making it easier to classify fake news. The study shows that advanced mixed models are better than old-fashioned ways. It also gives us a solid foundation for making more progress in this important area in the future. The study shows how important it is to choose the right model designs to deal with the complicated problem of fake news on social media.