Large Language Models na Identificação de Ideação Suicida em Textos não Clínicos em Inglês
DOI:
https://doi.org/10.5753/reic.2026.7642Keywords:
Aprendizagem de Máquina, Inteligência Artificial, Large Language Model, Ideação Suicida, Processamento de Linguagem NaturalAbstract
O suicídio é um fenômeno complexo que pode iniciar com ideação suicida e evoluir para planos e tentativas. Com a popularização das redes sociais, usuários frequentemente expressam sentimentos negativos nesses ambientes, possibilitando a identificação de sinais de risco. Este estudo tem como objetivo avaliar o desempenho de modelos baseados em Transformers e Large Language Models (LLMs) na detecção de ideação suicida em textos. A metodologia consistiu na aplicação e comparação de modelos discriminativos (ALBERT, Mental-RoBERTa e BERT-Large) e generativos (DeepSeek e ChatGPT-4o) em dois conjuntos de dados: ''Suicide and Depression Detection'' e ''Suicide vs Depression Classification'', utilizando métricas como acurácia, precisão, sensibilidade e Medida-F1. Os resultados indicaram alto desempenho dos modelos discriminativos no primeiro conjunto (até 0,99), enquanto os LLMs apresentaram melhor equilíbrio na detecção da classe suicide, com F1 de até 0,74. Esses achados evidenciam o potencial dessas abordagens para apoiar sistemas de saúde mental.
Downloads
Referências
Adadi, A. and Berrada, M. (2018). Peeking inside the black-box: A survey on explainable artificial intelligence (xai). IEEE Access, 6:52138–52160. DOI: 10.1109/ACCESS.2018.2870052.
Adadi, A. and Berrada, M. (2020). Explainable ai for healthcare: From black box to interpretable models. In Bhateja, V., Satapathy, S. C., and Satori, H., editors, Embedded Systems and Artificial Intelligence, pages 327–337, Singapore. Springer Singapore. DOI: 10.1007/978-981-15-0947-6_31.
Aladağ, A. E., Muderrisoglu, S., Akbas, N. B., Zahmacioglu, O., and Bingol, H. O. (2018). Detecting suicidal ideation on forums: Proof-of-concept study. J Med Internet Res, 20(6):e215. DOI: 10.2196/jmir.9840.
Arendt, F., Markiewitz, A., and Scherr, S. (2021). Investigating Suicide-Related Subliminal Messages on Instagram. Crisis, 42(4):263–269. DOI: 10.1027/0227-5910/a000717.
Bertolote, J. (2016). O suicídio e sua prevenção. Editora Unesp.
Bhaumik, R., Srivastava, V., Jalali, A., Ghosh, S., and Chandrasekharan, R. (2023). Mindwatch: A smart cloud-based ai solution for suicide ideation detection leveraging large language models. medRxiv. DOI: 10.1101/2023.09.25.23296062.
Birjali, M., Kasri, M., and Beni-Hssane, A. (2021). A comprehensive survey on sentiment analysis: Approaches, challenges and trends. Knowledge-Based Systems, 226:107134. DOI: 10.1016/j.knosys.2021.107134.
Braciszewski, J. M. (2021). Digital technology for suicide prevention. Advances in psychiatry and behavioral health, 1(1):53–65. DOI: 10.1016/j.ypsc.2021.05.008.
Calvo, R., Milne, D., Hussain, S., and Christensen, H. (2017). Natural language processing in mental health applications using non-clinical texts. Natural Language Engineering, pages 1–37. DOI: 10.1017/S1351324916000383.
Chadaga, K., Prabhu, S., Sampathila, N., Chadaga, R., Bhat, D., Sharma, A. K., and Swathi, K. (2024). SADXAI: Predicting social anxiety disorder using multiple interpretable artificial intelligence techniques. SLAS Technology, 29(2):100129. DOI: 10.1016/j.slast.2024.100129.
de Oliveira, A. C., Azevedo, J. P. C., Ruback, L., Moreira, R., Teixeira, S. S., and Teles, A. S. (2025). Effect of explainable artificial intelligence on trust of mental health professionals in an ai-based system for suicide prevention. IEEE Access, 13:60987–61005. DOI: 10.1109/ACCESS.2025.3556245.
de Oliveira, A. C., Diniz, E. J., Teixeira, S., and Teles, A. S. (2022). How can machine learning identify suicidal ideation from user’s texts? towards the explanation of the boamente system. Procedia Computer Science, 206:141–150. International Society for Research on Internet Interventions 11th Scientific Meeting. DOI: 10.1016/j.procs.2022.09.093.
Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. Disponível em: [link].
Diniz, E. J., Fontenele, J. E., de Oliveira, A. C., Bastos, V. H., Teixeira, S., Rabêlo, R. L., Calçada, D. B., Dos Santos, R. M., de Oliveira, A. K., and Teles, A. S. (2022). Boamente: a natural language processing-based digital phenotyping tool for smart monitoring of suicidal ideation. In Healthcare, volume 10, page 698. MDPI. DOI: 10.3390/healthcare10040698.
Fan, A., Jernite, Y., Perez, E., Grangier, D., Weston, J., and Auli, M. (2019). ELI5: long form question answering. CoRR, abs/1907.09190. Disponível em: [link].
Gholi Zadeh Kharrat, F., Gagne, C., Lesage, A., Gariépy, G., Pelletier, J.-F., Brousseau-Paradis, C., Rochette, L., Pelletier, E., Lévesque, P., Mohammed, M., et al. (2024). Explainable artificial intelligence models for predicting risk of suicide using health administrative data in quebec. PLoS one, 19(4):e0301117. DOI: 10.1371/journal.pone.0301117.
Gohel, P., Singh, P., and Mohanty, M. (2021). Explainable ai: current status and future directions. arXiv preprint arXiv:2107.07045. DOI: 10.48550/arXiv.2107.07045.
Haque, A., Reddi, V., and Giallanza, T. (2021a). Deep learning for suicide and depression identification with unsupervised label correction. In Farkaš, I., Masulli, P., Otte, S., and Wermter, S., editors, Artificial Neural Networks and Machine Learning – ICANN 2021, pages 436–447, Cham. Springer International Publishing. DOI: 10.1007/978-3-030-86383-8_35.
Haque, A., Reddi, V., and Giallanza, T. (2021b). SDCNL (Suicide vs Depression Classification). Disponível em: [link]. Acesso em 05 de junho de 2025.
Heston, T. F. and Khun, C. (2023). Prompt engineering in medical education. International Medical Education, 2(3):198–205. DOI: 10.3390/ime2030019.
Ji, S., Zhang, T., Ansari, L., Fu, J., Tiwari, P., and Cambria, E. (2022). MentalBERT: Publicly available pretrained language models for mental healthcare. In Calzolari, N., Béchet, F., Blache, P., Choukri, K., Cieri, C., Declerck, T., Goggi, S., Isahara, H., Maegaard, B., Mariani, J., Mazo, H., Odijk, J., and Piperidis, S., editors, Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 7184–7190, Marseille, France. European Language Resources Association. Disponível em: [link].
Komati, N. (2021). Suicide and depression detection. Repositório Kaggle. Disponível em: [link]. Acesso em 05 de junho de 2025.
Lan, Z., Chen, M., Goodman, S., Gimpel, K., Sharma, P., and Soricut, R. (2020). Albert: A lite bert for self-supervised learning of language representations. Disponível em: [link].
Lekkas, D., Klein, R. J., and Jacobson, N. C. (2021). Predicting acute suicidal ideation on instagram using ensemble machine learning models. Internet Interventions, 25:100424. DOI: 10.1016/j.invent.2021.100424.
Linardatos, P., Papastefanopoulos, V., and Kotsiantis, S. (2021). Explainable ai: A review of machine learning interpretability methods. Entropy, 23(1). DOI: 10.3390/e23010018.
Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., and Stoyanov, V. (2019). Roberta: A robustly optimized bert pretraining approach. Disponível em: [link].
Malhotra, A. and Jindal, R. (2024). Xai transformer based approach for interpreting depressed and suicidal user behavior on online social networks. Cognitive Systems Research, 84:101186. DOI: 10.1016/j.cogsys.2023.101186.
Molnar, C. (2020). Interpretable Machine Learning. Leanpub.
Neeleman, J., de Graaf, R., and Vollebergh, W. (2004). The suicidal process; prospective comparison between early and later stages. Journal of affective disorders, 82(1):43–52. DOI: 10.1016/j.jad.2003.09.005.
Nielsen, S. D., Christensen, R. H. B., Madsen, T., Karstoft, K.-I., Clemmensen, L., and Benros, M. E. (2023). Prediction models of suicide and non-fatal suicide attempt after discharge from a psychiatric inpatient stay: A machine learning approach on nationwide danish registers. Acta Psychiatrica Scandinavica, 148(6):525–537. DOI: 10.1111/acps.13629.
Nikhileswar, K., Vishal, D., Sphoorthi, L., and Fathimabi, S. (2021). Suicide ideation detection in social media forums. In 2021 2nd International Conference on Smart Electronics and Communication (ICOSEC), pages 1741–1747. DOI: 10.1109/ICOSEC51865.2021.9591887.
Nordin, N., Zainol, Z., Mohd Noor, M. H., and Chan, L. F. (2023). An explainable predictive model for suicide attempt risk using an ensemble learning and shapley additive explanations (shap) approach. Asian Journal of Psychiatry, 79:103316. DOI: 10.1016/j.ajp.2022.103316.
Oliveira, A. C. d., Bessa, R. F., and Teles, A. S. (2024). Comparative analysis of bert-based and generative large language models for detecting suicidal ideation: a performance evaluation study. Cadernos de Saúde Pública, 40(10):e00028824. DOI: 10.1590/0102-311XEN028824.
Oliveira, A. C. d., Diniz, E. J., Teixeira, S., and Teles, A. S. (2022). How can machine learning identify suicidal ideation from user’s texts? towards the explanation of the boamente system. Procedia Computer Science, 206:141–150. International Society for Research on Internet Interventions 11th Scientific Meeting. DOI: 10.1016/j.procs.2022.09.093.
Pal, A., Barigidad, A., and Mustafi, A. (2020). IMDb Movie Reviews Dataset. Disponível em: [link].
Pires, T., Schlinger, E., and Garrette, D. (2019). How multilingual is multilingual BERT? In Korhonen, A., Traum, D., and Màrquez, L., editors, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4996–5001, Florence, Italy. Association for Computational Linguistics. DOI: 10.18653/v1/P19-1493.
Ribeiro, M. T., Singh, S., and Guestrin, C. (2016). "why should i trust you?": Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’16, page 1135–1144, New York, NY, USA. Association for Computing Machinery. DOI: 10.1145/2939672.2939778.
Rizvi, A., Harmer, B., and Saadabadi, A. (2025). Suicidal Ideation. StatPearls Publishing, Treasure Island (FL). Disponível em: [link].
Rudin, C. (2019). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1(5):206–215. DOI: 10.1038/s42256-019-0048-x.
Safavian, S. and Landgrebe, D. (1991). A survey of decision tree classifier methodology. IEEE Transactions on Systems, Man, and Cybernetics, 21(3):660–674. DOI: 10.1109/21.97458.
Shatte, A. B. R., Hutchinson, D. M., and Teague, S. J. (2019). Machine learning in mental health: a scoping review of methods and applications. Psychological Medicine, 49(9):1426–1448. DOI: 10.1017/S0033291719000151.
Sinyor, M., Williams, M., Zaheer, R., Loureiro, R., Pirkis, J., Heisel, M. J., Schaffer, A., Redelmeier, D. A., Cheung, A. H., and Niederkrotenthaler, T. (2021). The association between twitter content and suicide. Australian & New Zealand Journal of Psychiatry, 55(3):268–276. PMID: 33153274. DOI: 10.1177/0004867420969805.
Sokolova, M. and Lapalme, G. (2009). A systematic analysis of performance measures for classification tasks. Information Processing Management, 45(4):427–437. DOI: 10.1016/j.ipm.2009.03.002.
Souza, V., Nobre, J., and Becker, K. (2021). A deep learning ensemble to classify anxiety, depression, and their comorbidity from texts of social networks. Journal of Information and Data Management, 12. DOI: 10.5753/jidm.2021.1901.
Tang, H., Miri Rekavandi, A., Rooprai, D., Dwivedi, G., Sanfilippo, F. M., Boussaid, F., and Bennamoun, M. (2024). Analysis and evaluation of explainable artificial intelligence on suicide risk assessment. Scientific Reports, 14(1):6163. DOI: 10.1038/s41598-024-53426-0.
Torous, J., Kiang, M. V., Lorme, J., Onnela, J.-P., et al. (2016). New tools for new research in psychiatry: a scalable and customizable platform to empower data driven smartphone research. JMIR mental health, 3(2):e5165. DOI: 10.2196/mental.5165.
Weisberg, S. (2005). Applied Linear Regression. Wiley Series in Probability and Statistics. Wiley.
Downloads
Published
Como Citar
Issue
Section
Licença
Copyright (c) 2026 Os autores
Este trabalho está licenciado sob uma licença Creative Commons Attribution 4.0 International License.
