Diffusion pattern-based misinformation discernment on social networks

Abstract

The advent of the internet and digitization of media have enabled billions of people around the world to overcome geographic barriers and have fast and easy access to information. However, it has also become an ideal place for the spread of misinformation. In order to decrease the exposure of users to misleading information, we aim to develop a robust model that predicts and controls the spread of fake news on online social media (e.g. Twitter) in an efficient and more robust fashion by taking into consideration only the diffusion patterns of misinformation. To do so, we developed several different detection algorithms based on graph neural networks (GNN), and only diffusion patterns were used to train the models. This promising technique deals with heterogeneous data and takes graphs as inputs. Our empirical results show a successful implementation of the models and enhancement of the accuracy. Apart from its technical contribution, this research can be a remarkable step towards harnessing information flow in online social media.