Vaccine Hesitancy in Discussion Forums: Computer-Assisted Argument Mining with Topic Models

"Arguments used when vaccination is debated on Internet discussion forums might give us valuable insights into reasons behind vaccine hesitancy."

This book chapter shares the results of a study that applied automatic topic modelling to a collection of 943 discussion posts in which vaccine was debated. The researchers undertook the investigation based on their conviction that more might be learned about the reasons for vaccine hesitancy by studying the arguments that are given in internet discussion forums for avoiding vaccination. To be able to learn from and monitor these discussions on a large scale, the content of large text collections needs to be coded. This is an intractable task when using manual coding approaches that require the entire text collection to be read, but is possible if important information could be automatically extracted and presented for manual coding.

The texts that were used for exploring computer-assisted coding were vaccine-related discussion threads from the British parental website Mumsnet, which hosts online forums where subjects related to parenting are discussed. The procedures of LDA (Latent Dirichlet Allocation) and NMF (Non-Negative Matrix Factorisation) were used as topic modelling methods.

Six distinct discussion topics were detected by the algorithm; arguments both for and against vaccination were identified for all the 6 topics analysed. When manually coding the posts ranked as most typical for these 6 topics, a set of semantically coherent arguments were identified for each extracted topic. Both Topic 1 and Topic 4 were related to MMR (measles, mumps, and rubella) vaccination, but the themes of Topic 1 were related to research on and reports of adverse vaccine reactions, while the Topic 4 posts discussed the duration of vaccine immunity, disease severity, and single vaccines. Topic 3 was related to the eradication of diseases through vaccinations, opinions on how smallpox was eradicated, and how that should affect vaccination programmes for other diseases. Topic 5 was related to trust or distrust in the medical profession and industry, as well as to attitudes towards vaccination within the medical profession. The arguments for Topic 6 revolved around risk assessments for child vaccination, for vaccine-preventable diseases, and for infecting vulnerable individuals.

The researchers conclude that this "indicates that topic modelling is a useful method for automatically identifying vaccine-related discussion topics and for identifying debate posts where these topics are discussed. This functionality could facilitate manual coding of salient arguments, and thereby form an important component in a system for computer-assisted coding of vaccine-related discussions."