A Review of Natural Language Processing for Social Media-Based Public Health Surveillance and Analytics

Social media platforms generate vast volumes of user-generated text reflecting public health attitudes, behaviours, misinformation exposure, and health event responses in near real time. This paper presents a comprehensive review of natural language processing approaches for social media-based public health surveillance across five domains: influenza and infectious disease surveillance, vaccine sentiment monitoring, mental health signal extraction, opioid crisis surveillance, and health policy discourse analysis. Lexicon-based methods including VADER and TextBlob, classical machine learning classifiers with TF-IDF feature representations, and transformer-based models including BioBERT and ClinicalBERT are systematically evaluated. Transformer-based models consistently outperform classical methods by 8 to 15 percentage points on domain-specific health text classification benchmarks. A responsible social media health surveillance framework integrating UK GDPR data governance and NHS information governance is proposed, with a comparative NLP performance table.