Using Artificial Neural Networks to Identify COVID-19 Misinformation

Since the spread of the coronavirus disease (COVID-19), a huge amount of information about the virus has been published over the internet and social networks. Along with such, there is an uncontrolled spread of harmful misinformation. This paper aims to review three state-of-the-art datasets of misinformation on COVID-19 and present experimental comparison on these datasets using various Neural Network architectures. The datasets comprise data from various sources such as articles from trusted websites and posts and tweets from social media. As for the algorithms, different NeuralNetwork architectures (ANN, CNN, RNN, and LSTM) are used to compare the reviewed datasets to detect misinformation about COVID-19. The experiments are conducted on the datasets individually andmerged together to generate models with larger input dataset. The results show, in terms of accuracy, that feedforward Artificial Neural Network (ANN) outperformed other more complicated Deep Learning methods such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). Moreover, merging the datasets has resulted in better performance in comparison to the individual datasets. In terms of execution time, ANN showed better performance with shorter training time.