COVID-19 fake news detection: A hybrid CNN-BiLSTM-AM model

Abstract

With the rapid development of technology, social media as a communication platform has caused a significant increase in the dissemination of false information and fake news. We propose an outlier knowledge management framework of “generation–spread–identification–refutation” for detecting fake news in emergencies based on the theory of complex adaptive systems and information transmission. We extract and acquire outlier knowledge of COVID-19 fake news, incorporate it into the outlier knowledge base of major emergencies for knowledge sharing and transformation, classify the acquired knowledge from three dimensions (people, organization, and technology), and develop wisdom according to the extracted knowledge. Our proposed hybrid model is based on Convolutional Neural Network, Bidirectional Long Short-term Memory Network, and Attention Mechanism (AM) for fake news detection, thereby improving the evaluation indicators of Loss, Accuracy, F1-score, and Recall by at least 1 %. The multi-head AM performs better in fake news detection when models with different AMs are adjusted, and significant differences in sentence length and topic distribution can be observed between real and fake news. The model provides a think tank and a platform for public opinion guidance to deal with major emergency news detection.