People’s attitudes toward vaccines can now be detected in their social media posts using a smart AI model, developed by researchers at the University of Warwick.
The AI-based model can analyze a post on social media and establish its author’s position toward vaccines, by being “trained” to recognize that position from a small number of example tweets.
As a simple example, if a publication contains mentions of mistrust in health care institutions, a fear of needles, or something related to a known conspiracy theory, the model may recognize that the person who wrote it is likely to feel negative. towards vaccinations.
Research funded by UK Research and Innovation (UKRI) will be presented today (July 12) at the 2022 Annual Conference of the American Chapter of the Association for Computational Linguistics.
It is led by Professor Yulan He of the University’s Computer Science Department, which is supported by a 5-year Turing AI Fellowship funded by the EPSRC.
Professor He and colleagues at the University of Warwick have used a data set of 1.9 million tweets in English, published from February to April 2021, to develop the vaccine attitude detection model (VADet).
VADet first analyzed the flow of tweets about COVID-19 vaccines, learning an increasing variety of elements and contexts relevant to the ongoing vaccination debate. The model then gradually reduced its analyzes by looking at the patterns that characterize users ’concerns and attitudes.
VADet looks for statistical patterns in words related to different topics or positions. It is built from a large-scale language model previously trained with a large amount of text from books in English and Wikipedia and has already acquired some language skills. He was then trained through vaccine-related tweets so that he would understand what topics have been addressed in these tweets.
The researchers manually tagged a small amount of these tweets with information about the user’s position regarding the topics covered in the vaccine-related tweets. VADet can take advantage of such a small number of tagged tweets to distinguish semantic information related to posture and topic from other unlabeled tweets.
The AI model then organized the tweets into groups of similar aspects, forming geometric patterns that visually demonstrate how certain views on vaccinations (pro-vaccination, anti-vaccination, or neutral) can be related to characteristics or specific detectable references in a social media post. .
The model could be used to provide information on why people are negative about vaccination, information that government and health organizations can use to design better-targeted messages to reassure the general public about vaccination.
The COVID pandemic intensifies the use of social media. People express their attitudes towards public health issues, including vaccinations against COVID-19. We have shown that it is possible to monitor social media traffic, detect attitudes against vaccines, and segment tweets into groups that discuss similar issues. This real-time monitoring of public attitudes could help health organizations and government agencies address vaccines and combat vaccine misinformation in a timely manner. “
Professor Yulan He of the Warwick Department of Computer Science and Artificial Intelligence Acceleration at the Alan Turing Institute
The key to progress lies in the specially developed algorithm, which has two crucial capabilities. First, you can leverage large-scale social media data about vaccination to detect issues automatically. This is done by inserting a thematic layer into an existing pre-trained language model.
Second, the algorithm can be adapted to a small set of social media posts tagged with vaccine attitudes to automatically detect particular patterns of topics and attitudes associated with topics. “This so-called adaptive self-improvement ability has not been previously explored for vaccine attitude detection,” says Lixing Zhu, a doctoral student in the Warwick Department of Computer Science who implemented the VADet model.
Professor He added: “The WHO identified doubt about the vaccine as one of the top ten health threads in the world in 2019. By automatically detecting vaccine attitudes on social media, our solution has the potential to allow for more timely intervention to address vaccination concerns. “