Machine Learning and NLP Approach to Predict Hospitalization Upon Adverse Drug Reaction Symptoms of Covid-19 Vaccine Administration

The COVID-19 pandemic has persisted for over one year and nine months. Identifying vaccines with higher-than-average rates of adverse reactions (ADRs) is crucial to take appropriate measures. This paper proposes leveraging machine learning and artificial intelligence advancements to develop a computer-based decision system to provide a more proactive approach to identifying and treating vaccine-associated ADRs, potentially mitigating hospital overcrowding during critical periods. To this end, we aim to create a predictive model that can identify the requirement for hospitalization by examining symptom notes of individuals who suffered from adverse reactions after receiving the COVID-19 vaccine. Our findings demonstrate that character and word embedding techniques offer superior results to domain-based embeddings (BioBERT) and SentenceBert. Implementing this model in hospitals could expedite decision-making for Covid-19 vaccine adverse reactions, potentially mitigating hospital overcrowding during critical period.