Detection of Propaganda and Bias in Social Media: A Case Study of the Israel-Gaza War (2023)

This study explores bias and propaganda detection in Arabic social media narratives surrounding the Israel-Gaza War (2023). Given the influence of biased content on public opinion and political discourse, we employ Natural Language Processing (NLP) techniques, Arabic BERT transformers, and traditional machine learning models, including SVM, Logistic Regression, and SGD. Using the Sina dataset from FigNews 2024 (October 7, 2023 - January 31, 2024), AraBERT achieves the highest accuracy for bias detection 74.00% with a precision of 68.00%, recall of 64.00%, and F1-score of 65.00%. For propaganda detection, SVM attains the highest accuracy 67.42% with a precision of 56.40%, recall of 51.51%, and an F1-score of 46.37%. The results highlight the impact of class imbalance on recall and F1-score, affecting overall performance. While BERT models excel in capturing linguistic nuances, traditional classifiers remain competitive for specific tasks. Future work will focus on mitigating class imbalance and exploring hybrid approaches to improve Arabic text classification in conflict-related discourse.