Amid ongoing developments in the Middle East, particularly in Gaza, media coverage and online discussions have reflected a wide range of perspectives on the conflict. Following October 7th, 2023, reporting and commentary on Gaza became a subject of significant debate. To examine this discourse from a linguistic perspective, especially on social media platforms such as Facebook, a group of authors (Zaghouani et al., 2024) introduced the FIGNEWS2024 dataset, which focuses on media bias in the early stages of the conflict. This study makes use of the Arabic and English portions of the dataset to fine-tune Transformer-based classifiers. To assess generalization, two Out-of-Domain datasets are also used as test sets. In addition, three large language models (LLMs) are employed for each language to paraphrase the original sentences as a form of data augmentation. The generated sentences are evaluated to determine whether they preserve the meaning of the originals. Results indicate that the synthetic sentences improve classifier performance on both the in-domain and out-of-domain datasets. Nevertheless, despite these performance gains, the paraphrased sentences often exhibit reduced realism, with outputs that are readily identifiable as AI-generated.