Triagem Emocional Urbana: Empatia e Segurança em Modelos Fundacionais

Karina Flor Condori-Bustincio; Eloisa Lizeth Paredes-Cayllahua; Aníbal Sardón Paniagua; Rómulo W. C. Bustincio; Wellington Viana Lobato Junior

doi:10.5753/courb.2026.24128

Karina Flor Condori-Bustincio Universidad Nacional de San Agustín de Arequipa https://orcid.org/0009-0004-5400-8136
Eloisa Lizeth Paredes-Cayllahua Universidad Nacional de San Agustín de Arequipa https://orcid.org/0000-0001-7561-6991
Aníbal Sardón Paniagua Universidad Tecnológica del Perú https://orcid.org/0000-0002-4155-9631
Rómulo W. C. Bustincio UNICAMP https://orcid.org/0000-0002-3971-5105
Wellington Viana Lobato Junior INRIA-Saclay https://orcid.org/0000-0003-3858-8578

DOI: https://doi.org/10.5753/courb.2026.24128

Resumo

A integração de Modelos de Linguagem de Grande Escala (LLMs) em infraestruturas de Cidades Inteligentes amplia o suporte emocional inicial, mas impõe desafios de segurança clínica e autenticidade. Este estudo estabelece o conceito de Triagem Emocional Urbana, voltado à avaliação de LLMs em quiosques autônomos para identificação de perfis de risco e apoio conversacional. A metodologia baseia-se em um pipeline de simulação com diálogos multi-turno (10–16 turnos) entre um simulador de utilizador e modelos fundacionais (GPT-5.4-mini e DeepSeek-Chat), auditados por um Juiz LLM. Avaliamos o trade-off entre métricas de segurança clínica — como acurácia diagnóstica e ausência de falsos negativos em casos severos — e indicadores de ciberpsicologia, como empatia percebida e sobre-empatia. Resultados preliminares indicam que, embora ambos os modelos preservem a segurança clínica, o DeepSeek-Chat apresenta maior consistência na adequação do tratamento e naturalidade discursiva. O estudo estabelece referenciais metodológicos para a implementação ética e clinicamente segura de sistemas de IA em redes urbanas de saúde mental.

Referências

Alhuzali, H. and Alasmari, A. (2024). Evaluating the effectiveness of the foundational models for q&a classification in mental health care. arXiv preprint arXiv:2406.15966.

Alshuwaikhat, H. M., Aina, Y. A., and Binsaedan, L. (2022). Analysis of the implementation of urban computing in smart cities: A framework for the transformation of saudi cities. Heliyon, 8(8):e11138.

Burrell, D. N. (2025). Exploring the cyber complexity and cyberpsychology of the internet of things and ai tools in healthcare organizations. Brazilian Journal of Business, 7(4):079.

Dai, Z. (2024). Applications and challenges of large language models in smart government - from technological advances to regulated applications. In 2024 3rd International Conference on Frontiers of Artificial Intelligence and Machine Learning (FAIML 2024), pages 1–6. ACM.

Jeon, S. and Kim, H.-G. (2025). A comparative evaluation of chain-of-thought-based prompt engineering techniques for medical question answering. Computers in Biology and Medicine, 196:110614.

Khinvasara, T., Shankar, A., and Wong, C. (2024). Robustness and reliability testing in healthcare using artificial intelligence. Asian Journal of Research in Computer Science, 17(7):103–118.

Medetalibeyoglu, A., Velichko, Y. S., Hart, E. M., and Bagci, U. (2024). Foundational artificial intelligence models and modern medical practice. BJR—Artificial Intelligence, 2(1):ubae018.

Pan, F., Huang, X., Bi, Y., Gao, Y., Ye, Y., and Wang, H. (2025). From tools to partners: How large language models are transforming urban planning. AI Open, 6:276–298.

Saglam, S., Uludag, V., Karaduman, Z. O., Arıcan, M., Yücel, M. O., and Dalaslan, R. E. (2025). Comparative evaluation of artificial intelligence models gpt-4 and gpt-3.5 in clinical decision-making in sports surgery and physiotherapy: a cross-sectional study. BMC Medical Informatics and Decision Making, 25:163.

Shen, J., DiPaola, D., Ali, S., Sap, M., Park, H. W., and Breazeal, C. (2024). Empathy toward artificial intelligence versus human experiences and the role of transparency in mental health and social support chatbot design: Comparative study. JMIR Ment Health, 11:e62679.

Son, S., Koo, S., Zi, E. H., Jang, J., and Lim, H. (2026). Evaluating over-empathizing in emotional support conversations: A user-centered framework. Expert Systems With Applications, 308:131059.

Xu, F., Zhang, J., Gao, C., Liu, P., Feng, J., and Li, Y. (2026). Towards a foundational platform for generative agents in simulated city environment. PLOS Complex Systems, 3(3):e0000093.

Yang, K., Ji, S., Zhang, T., Xie, Q., Kuang, Z., and Ananiadou, S. (2023). Towards interpretable mental health analysis with large language models. arXiv preprint arXiv:2304.03347.

Zhang, Y., Mohawk, C. N., Han, K., Tida, V. S., Li, M., and Hei, X. (2026). Mhdash: An online platform for benchmarking mental health–aware ai assistants. In IEEE Southe-astCon.