The Performance of Graph Neural Network in Detecting Fake News from Social Media Feeds

Abstract

Misinformation spread due to fake news can have an adverse effect on society and individuals. One of the primary sources through which fake news spreads is social media. Fake news detection in social media is critical and at the same time, it is challenging to solve as the articles are written to appear credible. The nature of deliberate writing makes it more challenging to recognize fake news based on only news content; therefore, it is challenging to detect fake news with only natural language processing (NLP). Adding the users’ activity history and other auxiliary information becomes essential. Hence, in recent years, graph neural networks (GNN) gained momentum in detecting fake news. In this paper, we analyze the performance of the GNN-based model on fake news detection from social media threads and compare them with a traditional machine learning model, LSTM. From our analysis, we can conclude that GNN based models can perform better than baseline LSTM in terms of accuracy, F1-Score, precision, and recall.