Detecting Fake News About Covid-19 on Small Datasets with Machine Learning Algorithms

Abstract

Nowadays the problem of fake news in social media is dramatically increasing, especially when it refers to fake news about Covid-19, as it is a recent and global problem. Because of this fact, it is important to have the ability to detect and delete such news immediately. In our research we concentrate our efforts on detecting fake news about Coronavirus on small datasets, using the Constraint-2021 corpus: the full dataset (10,700 messages) and the limited dataset (1,000 messages). We compare classical Machine Learning Algorithms (4 algorithms: Logistic Regression, Support Vectors Machine, Gradient Boosting, Random Forest) algorithms of classification from the Scikit-learn library, GMDH-Shell tool (2 algorithms: Combi and Neuro), and Deep Neural Network (LSTM model). The results show that GMDH algorithms outperform traditional Machine Learning Algorithms and are comparable with Neural Networks models results on the limited dataset.

Keywords

• fake news
• covid-19
• scikit learn
• gmdh shell