Feature Integration-based Residual Deep Learning Model for Fake News Detection using Multimodal Data Sources

The rapid growth of the modern digital world has supported to spread the news rapidly on the social media sectors including Weibo, Facebook, and Twitter. Unconfirmed news is frequently spread in multimedia content form such as video, audio, picture, or text. The propagation of certain false news cheats the public results in protests and generates issues. Fake news propagation has become a significant issue in modern data because of the huge utilization of social media entities. One of the important operations is false news identification, yet the difficulty of some languages makes the identification of fake news complex. The conventional approaches cannot gather many contextual and semantic features from the documents in a specific multilingual text corpus. To overcome such issues, a residual-based deep learning model is introduced for fake news detection using multimodal information. Firstly, multimodal information like text and images are gathered from publicly available data sources. Further, the input text is undergone for pre-processing and it is used for extracting the features. Here, the Bidirectional Encoder Representations from Transformers (BERT) is employed as feature extraction as F1. Similarly, the features from the raw image are achieved by applying the Visual Geometry Group (VGG16), designated as F2. Finally, these two resultant features are concatenated together and subjected to the Residual Recurrent Neural Network (Res-RNN) for detecting fake news. Hence, the proposed system is examined with different metrics and compared with various approaches. Thus, the results declare that the suggested system achieves higher accuracy in avoiding the spreading of fake news over the media.