This research investigates multilingual propaganda detection by employing transformer-based models, specifically mBERT, XLM-RoBERTa, and mT5. The study utilizes a balanced dataset from the BiasFigNews corpus, annotated for propaganda and bias across five languages. The models were finely tuned to generate embeddings for classification tasks. The evaluation revealed mT5 as the most effective model, achieving an accuracy of 99.61% and an F1-score of 0.9961, followed by mBERT and XLM-RoBERTa with accuracies of 92% and 91.41%, respectively. The findings demonstrate the efficacy of transformer-based embeddings in detecting propaganda while also highlighting challenges in subtle class distinctions. Future work aims to enhance cross-lingual adaptability and explore lightweight models for resource-constrained settings.

Bias in news reporting significantly influences public perception, particularly in sensitive and polarized contexts like the Israel-Gaza conflict. Detecting bias in such cases presents unique challenges due to political, cultural, and ideological complexities, often amplifying disparities in reporting. While prior research has addressed media bias and dataset fairness, these approaches inadequately capture the nuanced dynamics of the Israel-Gaza conflict. To address this gap, we propose an NLP-based framework that leverages Nakba narratives as linguistic resources for bias detection in news coverage. Using a multilingual corpus focusing on Arabic texts, we apply rigorous data cleaning, pre-processing, and methods to mitigate imbalanced class distributions that could skew classification outcomes. Our study explores various approaches, including Machine Learning (ML), Deep Learning (DL), Transformer-based architectures, and generative models. The findings demonstrate promising advancements in automating bias detection, and enhancing fairness and accuracy in politically sensitive reporting.