Machine Learning Applications in Political Disinformation Detection

Abstract

Political disinformation practices in the modern digital cultures have become a dangerous issue that undermines the population and influences the political process.  This work makes use of a mixed-methods paradigm combining quantitative machine learning with qualitative content analysis to study how machine learning-based tools can be used to detect such deception.  Pretextual processes that were used to collect data included text normalization, stopword removal, and metadata extraction of the internet forums, news sources, and social media platforms.  The machine learning models of Random Forest, XGBoost, and Neural Networks were employed in categorizing disinformation and measured accuracy, precision, recall, and F1 score to measure the performance of each model.  The results indicate that machine learning can potentially identify misinformation, and some of them perform well in terms of robustness and classification.  Introduction of metadata, i.e., user information and dates of publication, to provide context and time trends enhanced model performance substantially.  Besides contributing to the theoretical and practical contributions to the field, this study provides valuable information regarding the manner in which machine learning can be applied to combat political disinformation.