ARCNN framework for multimodal infodemic detection. (February 2022)

• Record Type: Journal Article
• Authors: Raj, Chahat; Meel, Priyanka
• Is Part Of: Neural networks. Volume 146(2022)
• Page Start: 36
• Page End: 68
• Publication Date: 2022-02
• Subjects: COVID-19 fake news -- Infodemic -- Deep learning -- Multimodal fusion -- Neural networks; Neural computers -- Periodicals; Neural networks (Computer science) -- Periodicals; Neural networks (Neurobiology) -- Periodicals; Nervous System -- Periodicals; Ordinateurs neuronaux -- Périodiques; Réseaux neuronaux (Informatique) -- Périodiques; Réseaux neuronaux (Neurobiologie) -- Périodiques; Neural computers; Neural networks (Computer science); Neural networks (Neurobiology); Periodicals; 006.32
• DOI: 10.1016/j.neunet.2021.11.006
• Languages: English
• ISSNs: 0893-6080

Abstract

Abstract: Fake news and misinformation have adopted various propagation media over time, nowadays spreading predominantly through online social networks. During the ongoing COVID-19 pandemic, false information is affecting human life in many spheres The world needs automated detection technology and efforts are being made to meet this requirement with the use of artificial intelligence. Neural network detection mechanisms are robust and durable and hence are used extensively in fake news detection. Deep learning algorithms demonstrate efficiency when they are provided with a large amount of training data. Given the scarcity of relevant fake news datasets, we built the Coronavirus Infodemic Dataset (CovID), which contains fake news posts and articles related to coronavirus. This paper presents a novel framework, the Allied Recurrent and Convolutional Neural Network (ARCNN), to detect fake news based on two different modalities: text and image. Our approach uses recurrent neural networks (RNNs) and convolutional neural networks (CNNs) and combines both streams to generate a final prediction. We present extensive research on various popular RNN and CNN models and their performance on six coronavirus-specific fake news datasets. To exhaustively analyze performance, we present experimentation performed and results obtained by combining both modalities using early fusion and four types of late fusion techniques. The proposed framework is validated by comparisons with state-of-the-art fake news detection mechanisms, and our models outperform each of them. Highlights: A new multimodal Coronavirus Infodemic Dataset 'CovID' is introduced. An effective neural network ARCNN is proposed for multimodal fake news classification. Experimented with different variants of early and late fusion. Analysis is done on six neural networks. Experimented on six multimodal COVID-19 fake news datasets.