ABSTRACT
Misinformation has become a frightening specter of society, especially fake news that concerning Covid-19. It massively spreads on the Internet, and then induces misunderstandings of information to the national and global communities during the pandemic. Detecting massive misinformation on the Internet is crucial and challenging because humans have struggled against this phenomenon for a long time. Our research concerns detecting fake news related to covid-19 using augmentation [random deletion (RD), random insertion (RI), random swap (RS), synonym replacement (SR)] and several graph neural network [graph convolutional network (GCN), graph attention network (GAT), and GraphSAGE (SAmple and aggreGatE)] model. We constructed nodes and edges in the graph, word-word node, and word-document node to graph neural network. Then, we tested those models in different amounts of sample training data to obtain accuracy for each model and compared them. For our fake news detection task, we found training accuracy steadily increasing for GCN, GAT, and SAGE models from the beginning to the end of the epochs. This result proved that the performance of GNN, whether GCN, GAT, or SAGE gained an entirely insignificant difference precision result.
