In this study, we explore the semantics and linguistic features of vaccine hesitancy. To this end, we collected 2,000 tweets from the X platform (formerly Twitter). We applied a supervised technique, logistic regression, to identify the semantic features that predict vaccine skepticism. Next, we employed an unsupervised algorithm, Latent Dirichlet Allocation (LDA), for topic modeling and interpreted the top ten extracted topics. Our main results demonstrate the practical utility of text mining as a decision-support tool for public health professionals. By systematically analyzing social media discourse, policymakers and healthcare administrators can track emerging concerns in real time, prioritize resource allocation, and fine-tune public health campaigns based on evolving public sentiment. Keywords Vaccine, COVID-19, social media, text mining.