Fake News Detection Model with Hybrid Features—News Text, Image, and Social Context

Abstract

With the evolving realm of news propagation and the surge in social media usage, detecting and combatting fake news has become an increasingly important issue. Currently, fake news detection employs three main feature categories: news text, social context, and news images. However, most studies emphasize just one, while only a limited number incorporate image features. This study presents an innovative hybrid fake news detection model amalgamating text mining technology to extract news text features, user information on Twitter to extract social context features, and VGG19 model to extract news image features to increase the model's accuracy. We harness four diverse machine learning algorithms (Logistic Regression, Random Forest, Support Vector Machine, and Extreme Gradient Boosting) to construct models and evaluate their performance via Precision, Recall, F1-Score, and Accuracy metrics. Results indicate the fusion of news text, social context, and image features outperforms their individual application, yielding a noteworthy 92.5% overall accuracy. Significantly, social context attributes, encompassing users, publishers, and distribution networks, contribute crucial insights into detecting early-stage fake news dissemination. Consequently, our study bolsters fact-checking entities by furnishing them with news-content insights for verification and equips social media platforms with a potent fake news detection model—comprising news content, imagery, and user-centric social context data—to discern erroneous information.