Sentinela: protótipo para detecção de grooming em conversas online utilizando redes neurais recorrentes

Abstract

The growth of digital communications has increased the vulnerability of minors to the grooming process. The objective of this work is to develop a software prototype focused on the proactive detection of predatory behavior in online conversations. The quantitative methodology used real interaction data to train and compare machine learning algorithms, emphasizing Long Short-Term Memory (LSTM) recurrent neural networks capable of retaining the temporal context of dialogues. The validated model outperformed traditional approaches, achieving an 83.0% Recall and demonstrating high sensitivity in capturing threats. The artificial intelligence was encapsulated in a moderation interface, whose practical evaluation confirmed the technical viability and usefulness of the tool as a preventive protection mechanism.