Predicting Opinions in Social Networks Using Recurrent Neural Networks

Abstract

This paper studies the spread of opinions in social media networks through the lens of opinion dynamics. As more human interactions and public discourse move online, understanding opinion formation and evolution in social media is crucial for issues such as virtual marketing, information dissemination, and social security. We introduce a novel approach using recurrent neural networks (RNN) to monitor and predict interactions in these networks. Our method uses two configurations of RNN algorithms to predict the opinions of agents in an online social network, with results showing its effectiveness in predicting diverse opinions. The first configuration uses a sigmoid activation function to predict the binary opinions output (agree, disagree), while the second configuration uses the softmax function to predict more detailed opinions. For the simulation results, we considered a group of five agents interacting in the Twitter network on the subject of COVID-19. The social interaction for a 30-day period was captured and opinion dynamics prediction using the RNN was verified.