FakeNED: A Deep Learning Based-System for Fake News Detection from Social Media

Social networks are increasingly present in our daily life. They allow us to remain in contact with friends regardless of distances, to share posts, images, videos, to be part of communities or come across articles and news. Everything develops so quickly that formality and content are no longer given too much importance. Therefore, it is through the development of these social networks that the need to distinguish fake news from real ones has developed. In this paper, we propose FakeNED a system for the detection of fake news on social networks. It comprises a multimodal Deep Learning (DL) approach of extracting features both from the text and from the images of the article. For the first extraction, we implemented a BERT-based (Bidirectional Encoder Representations from Transformers) method with an initial pre-trained phase followed by a fine-tuning final phase. For the latter, we used a VGG-16 to develop the image feature extraction. The extracted features were then given as input to a Fully Connected Layer in order to obtain the final output. We conduct our experiments on Fakeddit dataset through which we obtained a result which outperforms the state-of-art models. Moreover, FakeNED includes a service for allowing users to easily estimate the truth of social media content.