Vaccine Hesitancy Classifier

This service classifies documents based on the stance towards COVID-19 vaccines expressed within the text.

The classifier was trained using the VaxxHesitancy dataset we released. Full details of the dataset can be found in the associated ICWSM 2023 paper.

Given a document, the service assigns one of the following four classes:

1. "Pro-vaccine": The tweet expresses opinions and actions supporting COVID-19 vaccination use. For example,
   1. The tweet delivers positive information about the COVID-19 vaccine (example: Coronavirus vaccine developed by Oxford University appears safe and trains the immune system. )
   2. The tweet debunks misinformation that potentially results in vaccine hesitancy (example: That COVID-19 vaccines were developed w fetal tissue? (No, they are not))
   3. The tweet shows clear support for the COVID-19 vaccine (example: Happy to have received my jab today! Thanks NHS!)
   4. The tweet encourages others to get vaccinated (example: Just get jabbed guys)
2. "Anti-vaccine": The tweet expresses opinions and actions against COVID-19 vaccination, with the aim of persuading others to refuse vaccination. Please be aware that tweets expressing the person's own intention to refuse vaccination themselves belong to the "Vaccine Hesitant" category. Examples of anti-vaccine tweets include tweets that:
   1. Discourage others from getting vaccinated (example: Giving people the 'choice' to be vaccinated and then limiting those who chose not to, access to service is wrong and is setting a stage for medical tyranny. #mybodymychoice)
   2. Spread (mis)information that potentially results in vaccine hesitancy (example: Will it alter DNA & remote control ppl thru 5g ? Yes)
   3. Advise people how to decline a vaccine (example: How to (possibly) decline a vaccine: Decline the #vaccine on moral, religious or medical grounds)
3. "Vaccine Hesitant": The tweet is centred on the person's intention to delay and/or refuse the vaccine and is often in first person. For example,
   1. Concern about side effects, particularly unknown long-term side effects (example: And you hear all these horror stories about things that are going wrong with you...and just like the blood clots and people like being paralysed and like the part of the body is gone numb and things like that)
   2. Rushed vaccine development (example: I think it came out about 8 months after the pandemic hit and I was like absolutely no, that's too quick. I would not be happy with that.
   3. Distrust of the government (example: I believe there's an element of truth in the fact that Governments do want to control and have more access to people’s... data and everything like that. It's important to them)
   4. Not knowing what's in the vaccine (example: I can not take the #vaccine since it's made of #aborted fetuses, how about you?)
   5. Asks / @mentions other users about the COVID-19 vaccine (tweet: What are the side-effects of COVID vaccine??)
4. "Irrelevant": Any tweet that does not express a stance towards the COVID-19 vaccine, e.g. tweets about other kinds of vaccines or tweets primarily about COVID-19 or other aspects of the pandemic.

The maximum input document length is roughly 512 tokens due to the nature of BERT's input limits. This means that any text exceeding the first 512 tokens will be truncated by the tokenizer.

This research was supported by a UKRI grant EP/W011212/1 (XAIvsDisinfo: eXplainable AI Methods for Categorisation and Analysis of COVID-19 Vaccine Disinformation and Online Debates) and an EU Horizon 2020 grant (agreement no.871042 -- "So-BigData++: European Integrated Infrastructure for Social Mining and BigData Analytics").

Annotation details