Hybrid deep learning information integrated model for fake news detection in social media

Abstract

Digital media forms like blogs, online news media, and social media occupied the places of earlier news transmission platforms like newspapers and magazines. Inter linked referrals of these platforms help the news to be spread within no time. This fascination leads to invading negativity like fake news and differently portrayed information. Fake information negatively affects all the fields like health, education, government, and the market because people make decisions about anything based on the information available. The news may be text-based or multi-modal based. Any combination of text, image, and video may be present in multi-modal news. Attention seekers create fake news by altering text, images, or both. The spread of fake news is in different domains, but fact-checking websites can check the authenticity of a particular environment; hence, fake news detection remains challenging. Another reason behind the difficulty in fake news detection is the unstructured representation (in the form of articles, pictures, audio, video, etc.) of the news, which needs a human to classify. In spite of plenty of research work which has been done for meeting this purpose, proper classification still faces various challenges like imbalance, multi-modality, lack of appropriate structure, and ambiguity of words in the datasets. This research proposes four novel deep learning architectures for fake news classification based on text and image. Titles explains news headlines and captures anyone s attention towards some information. The image provides relevant pictorial data about the news. In much fake news, the text content and visual information will not have relationship. This work is a hybrid model which utilizes all the three pieces of information namely title, text and image. In the first work, the Deep-Learned Bidirectional Gated Recurrent Unit (Bi GRU)-Long Short-Term Memory (LSTM) Model (DL-BGLM) is used for detecting fake news using textual content and news title. newline