Social media language analysis for early detection of mental health disorders

• Grant number: 21/08213-0
• Support Opportunities: Regular Research Grants
• Duration: May 01, 2022 - April 30, 2024
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Ivandre Paraboni
• Grantee: Ivandre Paraboni
• Host Institution: Escola de Artes, Ciências e Humanidades (EACH). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated researchers: Deusivania Vieira da Silva Falcão ; Thiago Alexandre Salgueiro Pardo

Abstract

The observation that individuals with mental health disorders such as depression and anxiety are often regular users of social media has led to the development of a wide range of studies in the Natural Language Processing (NLP) field for risk assessment based on the kind of language employed by these individuals. Existing work in the field is however largely dedicated to the English language, and tends to consider publications (e.g., tweets) produced at any time, including even those produced after the individual is already clinically diagnosed. Thus, models of this kind tend to focus more on the issue of distinguishing individuals with and without a certain disorder, but are less able to anticipate these as a means to prevent their possible aggravation. Based on these observations, this project proposes to explore the temporal information provided by the Twitter platform for the study and development of computational models for early recognition of depression and anxiety disorder in Portuguese using a database - called the SeptemberBR corpus - carefully designed so as to include only texts that are chronologically prior to the date of diagnosis reported by social media users. A study of this kind, in addition to introducing a novel (and possibly more useful) formulation of the present computational problem, opens up the opportunity for a number of scientific contributions in the NLP field, including the modeling of textual and non-textual features and the use of recent neural learning methods, and enables novel solutions for a pressing issue of great social interest. (AU)