Deep Neural Architecture for Veracity Analysis of Multimodal Online Information

The extensive digital communication among users around the world in various fields such as politics, sports, religious beliefs, product promotions, celebrities, etc. has given rise to a huge volume of unauthenticated and unverified information. So, it has become imperative in the current scenario to design methods for veracity analysis of online information in various data formats mainly text and images. This work introduces a novel automatic fake news detection framework by utilizing the capabilities of Deep Neural Networks namely VGG16 for Image, Bi-LSTM for Text and Title part of online news. The designed model is intended to be robust enough for handling the forgeries present in the online news articles. Finally, the three parallel architectures for each part of the news are combined using the late fusion approach by some sets of normalized weights. Effective experimental analysis has been carried out on two different real-world news datasets which comprise of real and fake news articles. The highest classification accuracy achieved is 97.74% on the All Data dataset. The state-of-the-art comparison advocates the advantages of the proposed model over other contemporary techniques.