Detecting Signs of Depression in Portuguese Texts Considering Brazil’s Cultural and Regional Aspects Using LLMs

• Leidiane Rodrigues UFG
• Deborah Fernandes UFG
• Marilúcia Lago UFG

DOI: https://doi.org/10.5753/sbcas_estendido.2025.7765

Abstract

This study investigates the identification of depression traits in messages from the social network Reddit, focusing on young people and adolescents. To this end, two lexicons were compared: one with terms in Portuguese without cultural expressions and another culturally enriched lexicon, constructed from Brazilian song lyrics. Using the GPT 3.4 and Flash 1.5 models, we analyzed messages labeled as depressive and non-depressive, classified by a mental health specialist. The results show that the cultural adaptation of the lexicon allowed a more sensitive and accurate analysis, suggesting the importance of regional lexicons to improve the detection of signs of depression in specific cultural contexts.

References

Canaltech (2024). O que são prompts no chatgpt? Acessado em: 01 março de 2025.

FIA (2024). O que é chatgpt? Acessado em: 01 março de 2025.

Mendes, A.R. e Caseli, H. d. M. (2024). Detecção automática de postagens possivelmente depressivas em redes sociais. Acessado em: 01 maio 2024.

Reddit (2024). Reddit. Acessado em: 01 fevereiro de 2025.

Spitzer, R. L., Kroenke, K., Williams, J. B. W., and the Patient Health Questionnaire Primary Care Study Group (1999). Validation and utility of a self-report version of prime-md: The phq primary care study. Journal of the American Medical Association, 282(18):1737–1744.

Spotify (2025). Spotify - plataforma de streaming de música. Acesso em: 16 mar. 2025.