From Bag-of-Words to Reasoning: Comparing Traditional Supervised ML and Zero-Shot LLMs for Sexual Predator Identification in Brazilian Portuguese

• Leonardo Ferreira dos Santos CEFET/RJ
• Gustavo Guedes CEFET/RJ

DOI: https://doi.org/10.5753/brasnam.2026.23676

Resumo

Automated detection of online sexual predators has traditionally relied mostly on supervised classifiers using both textual and engineered features and annotation labels, simplifying the complexities of sexual predatory behavior. With the emergence of LLMs and the scarcity of real-world data, it is important to explore their potential in this research domain. In this work, four commercial LLMs with reasoning capabilities are evaluated in zero-shot mode on the PREDADORES-BR dataset for binary classification of predatory conversations in Brazilian Portuguese. The best-performing model achieved F₁ = 96% with 100% precision and zero false positives, with recall statistically indistinguishable from the best supervised baseline (SVM, F₁ = 89.87%).

Referências

Borj, P. R., Raja, K., and Bours, P. (2023). Online grooming detection: A comprehensive survey of child exploitation in chat logs. Knowledge-Based Systems, 259:110039.

Comitê Gestor da Internet no Brasil (2025). Pesquisa sobre o uso da Internet por crianças e adolescentes no Brasil: TIC Kids Online Brasil 2025. CGI.br, São Paulo.

dos Santos, L. and Guedes, G. (2019). Identificação de predadores sexuais brasileiros por meio de análise de conversas realizadas na internet. In Anais do VIII Brazilian Workshop on Social Network Analysis and Mining, pages 143–154, Porto Alegre, RS, Brasil. SBC.

dos Santos, L. F. and Guedes, G. P. (2018). Detecção de traços de narcisismo em conversas com predadores sexuais. In Anais do VII Brazilian Workshop on Social Network Analysis and Mining, pages 217–222, Porto Alegre, RS, Brasil. SBC.

Ebrahimi, M., Suen, C. Y., and Ormandjieva, O. (2016). Detecting predatory conversations in social media by deep convolutional neural networks. Digital Investigation, 18:33–49.

Hamm, L. and McKeever, S. (2025). Comparing machine learning models with a focus on tone in grooming chat logs. Frontiers in Pediatrics, 13:1591828.

Inches, G. and Crestani, F. (2012). Overview of the international sexual predator identification competition at pan-2012. CLEF (Online working notes/labs/workshop), 30.

Kloess, J. A., Hamilton-Giachritsis, C. E., and Beech, A. R. (2019). Offense processes of online sexual grooming and abuse of children via internet communication platforms. Sexual Abuse, 31(1):73–96.

Landis, J. R. and Koch, G. G. (1977). The measurement of observer agreement for categorical data. Biometrics, 33(1):159–174.

Lundberg, S. M. and Lee, S.-I. (2017). A unified approach to interpreting model predictions. In Advances in Neural Information Processing Systems, volume 30, pages 4765–4774.

Milon-Flores, D. F. and Cordeiro, R. L. F. (2022). How to take advantage of behavioral features for the early detection of grooming in online conversations. Knowledge-Based Systems, 240:108017.

Nguyen, T. T., Wilson, C., and Dalins, J. (2024). Fine-tuning llama 2 large language models for detecting online sexual predatory chats and abusive texts. In ESANN 2024 Proceedings, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, pages 613–618, Bruges, Belgium.

Olson, L. N., Daggs, J. L., Ellevold, B. L., and Rogers, T. K. (2007). Entrapping the innocent: Toward a theory of child sexual predators’ luring communication. Communication Theory, 17(3):231–251.

Panzariello, M. R. (2022). Estratégias para detecção precoce de predadores sexuais em conversas realizadas na internet. Dissertação de mestrado, COPPE, Universidade Federal do Rio de Janeiro, Rio de Janeiro.

Pendar, N. (2007). Toward spotting the pedophile telling victim from predator in text chats. International Conference on Semantic Computing (ICSC 2007), 1:235–241.

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, pages 1135–1144.

Santos, L. F. and Guedes, G. P. (2020). Identificação de predadores sexuais brasileiros em conversas textuais na internet por meio de aprendizagem de máquina. iSys: Revista Brasileira de Sistemas de Informação, 13(2):26–53.

Schenker, N. and Gentleman, J. F. (2001). On judging the significance of differences by examining the overlap between confidence intervals. The American Statistician, 55(3):182–186.

Villatoro-Tello, E., Juárez-González, A., Escalante, H. J., Montes-y Gómez, M., and Pineda, L. V. (2012). A two-step approach for effective detection of misbehaving users in chats. In CLEF (Online Working Notes/Labs/Workshop), volume 1178.