Cognitive COVID-19 Fake News Detection Model based on Machine Learning Approach

In today's era of information explosion, when information and knowledge are transmitted through social platforms, people often misbelieve wrong information or false information maliciously created, causing varying degrees of impact on society. This process is called "Infodemic". The term "information epidemic"first appeared during the SARS epidemic in 2003. False information spread rapidly and massively around the world through various communication channels, causing national security, economy, and politics to be affected. Therefore, this research applies the latent dirichlet allocation (LDA) method into the topic model, combined with TF and TF-IDF for COVID-19 fake news detection comparison. As the result of five classification models comparison - SVM, random forest, XGBoost and AdaBoost, the LDA combined with TF-IDF features can improve both SVM and random forest models of F1-score, among which the SVM model has the most significant improvement effect. After 10-fold cross-validation, the average F1-score growth rate of SVM increased by 1.13%, the accuracy was 98.04%, and the F1-score reached 98.10%.