RespiratoryPH: Empowering Low-Resourced Languages Through Multilingual and Multi-Labeled Social Media Dataset Towards Intelligent Public Health Disease Surveillance

The RespiratoryPH, a Gold Standard for a multilingual and multi-labeled social media dataset was developed aimed at enhancing public health surveillance for respiratory diseases in low-resourced languages. Data were collected from four major social media platforms—Facebook, Twitter, Reddit, and TikTok—utilizing a combination of API resources and web scraping tools, resulting in a dataset of 14,081 posts in English, Filipino, and Cebuano. Posts were manually annotated by two groups of medical experts for symptoms related to Acute Upper Respiratory Infection (AURI), Pneumonia, Pulmonary Tuberculosis, and COVID-19. The annotation process achieved a fair inter-rater reliability score (Scott's Pi = 0.3372). Several machine learning models used this dataset for training and testing, which achieved an accuracy of 81% for multilabel classification, demonstrating its potential for intelligent disease surveillance. The study highlights the importance of multilingual datasets in addressing public health challenges in linguistically diverse regions and underscores the need for continued research in lowresourced language processing.