Fake news in health: the validation of a neural network model for detection of disinformation in Pediatric Dentistry through psychophysiological reactions of internet users

• Grant number: 19/27242-0
• Support type: Scholarships in Brazil - Doctorate
• Effective date (Start): January 01, 2021
• Effective date (End): April 30, 2024
• Field of knowledge: Health Sciences - Dentistry - Pediatric Dentistry
• Principal researcher: Thiago Cruvinel da Silva
• Grantee: Matheus Lotto de Almeida Souza
• Home Institution: Faculdade de Odontologia de Bauru (FOB). Universidade de São Paulo (USP). Bauru , SP, Brazil

Abstract

The diffusion of information and communication technologies stimulate the production of contents based on personal opinions, contributing to the electronic dissemination of fake news via social media. In this sense, the consumption of disinformation directly influences the onset of beliefs that impact on people's health behaviors negatively. Hence, the development of tools for aiding Internet users to obtain and use health information properly is desirable. This study aims to develop and validate a machine learning model to detect the consumption of children's oral health-related disinformation remotely, based on psychophysiological reactions of mothers. The methods will be conducted in four distinct steps: 1) the determination of the interests of Google users in children's oral health-related disinformation, employing Google Trends; 2) the detection of children's oral health-related disinformation posted by Twitter users; 3) the assessment of differences of patterns in psychophysiological reactions of mothers exposed or not to disinformation, conducting a double-blind, cross-sectional and parallel randomized controlled trial; and 4) the training and evaluation of artificial neural networks to measure fake news-related psychophysiological outcomes, based on Cross Validation (CV) and Leave-One-Session-Out Cross-Validation (LOSOCV). Statistical analysis will be performed according to normality and homogeneity of data. P<0.05 will be considered for significant differences. (AU)