Detecting Fake News Spreaders on Social Media Using Posts Content Versus Profile Information

The rapid spread of misinformation on social media platforms poses significant societal challenges, underscoring the need for effective methods to detect fake news spreaders. While many studies combine textual and user profile features for this task, the individual impact of these attributes remains unclear, raising questions about how each feature independently influences model performance. This study evaluates the effectiveness of textual data and user profile information in detecting fake news spreaders. By utilizing the Bidirectional Long Short-Term Memory (BiLSTM) model on the TruthSeeker dataset, this study assesses the individual contribution of each feature and analyzes the effect of combining them on model accuracy. Additionally, the role of optimization techniques, such as L2 regularization and early stopping, in improving detection performance was investigated. The results indicate that textual data alone is highly effective (achieving an accuracy of 97.78%) in distinguishing between real and fake news, while the inclusion of user profile data (97.61%) does not significantly enhance model performance. Furthermore, models using only user profile data exhibited accuracy between 55% and 64%, highlighting the limited predictive power of this feature set alone.