This paper presents our contribution to Task 5 of the Social Media Mining for Health (#SMM4H) 2025 shared task, which addresses the classification of foodborne disease outbreaks and recall events from news articles, with a primary focus on FDA press releases. Task 5 is divided into two subtasks: (1) sentence-level classification of news articles, and (2) extraction of key entities and events from text. To tackle these challenges, we employed transformer-based architectures and large language models, leveraging both few-shot and zero-shot prompt engineering strategies. Additionally, we explored various prompt formulation and data augmentation techniques to assess their influence on system performance. Our proposed systems achieved an F1 score of 92.3% in Subtask 1 and an average score of 48% in Subtask 2 on the test dataset. On the test set, our models attained the second-highest accuracy in both subtasks.

PDF