Vaccination has been crucial in public health for decades, effectively controlling and eliminating many infectious diseases. However, COVID-19 pandemic has worsened the division in vaccine discussions on social media platforms, highlighting the profound impact of the crisis on both real-world and online interactions. This study aims to analyze and understand these discussions by creating a comprehensive vaccine dataset. The dataset includes 4,781 tweets and 10,889 quadruples that cover stance holders, vaccine entities, stance descriptions, and stance polarity labels. We utilize this dataset to timely identify people’s stances on social media, leveraging the chain-of-thought-based VSQ method to uncover relationships between stance holders and targets in tweets, extract their descriptions, and obtain stance information, this approach aids public health efforts in combating misinformation and promotes an accurate understanding of vaccine-related information. This work provides valuable resources for researchers and public health officials to better understand and address the polarized discourse surrounding vaccines. Our contributions include a novel publicly available dataset for vaccine stance detection, an LLM-CoT stance detection model, and substantial experiments beyond our dataset. Our code are available at https://github.com/shengdan320/VSDQ .