Psychological Reactance to Vaccine Mandates on Twitter: A Study of Sentiments in the United States

Hsieh, PH. Psychological reactance to vaccine mandates on Twitter: a study of sentiments in the United States. J Public Health Pol 46, 269–283 (2025). https://doi.org/10.1057/s41271-025-00554-0

The fine-tuned model for classifying tweets related to vaccine mandates is available on Hugging Face phsieh/vaccine-mandate-bertweet-classifier.

This repository contains the following files:

• training_no_texts.csv: Dataset with human-annotated labels used for fine-tuning classifiers to identify tweets related to vaccine mandates.

  • index: Original index from the annotated dataset before shuffling for training/testing splits.
  • id: Tweet ID.
  • mandate_m: Human-annotated label indicating whether a tweet is related to vaccine mandates (1 = related, 0 = not related).
  • usage: Indicates whether the tweet was used for training, evaluation, or testing.
  • pred: Predicted label from the fine-tuned model.
  • score0: Model score for the "not related to vaccine mandates" class. Can be transformed into a probability using the softmax function.
  • score1: Model score for the "related to vaccine mandates" class. Can be transformed into a probability using the softmax function.

• Mandate_tweets_Bertweet_training.ipynb: Code for fine-tuning BERTweet to classify tweets about vaccine mandates. Originally run on Google Colab.

• Mandate_tweets_supervised_learning.ipynb: Code for training a machine learning model using a bag-of-words approach to classify tweets. Originally run on Google Colab.

• Bertweet_inference.py: Code for classifying tweets using bertweet_mandate_classifier. Originally run on high-performance computing clusters.

• Vax_Dictionary_coding.py: Code for dictionary-based tweet classification. Originally run on high-performance computing clusters.

• Vaccine_roBERTa.py: Code for using TweetNLP’s sentiment and emotion classifiers. Originally run on high-performance computing clusters.