SMARTSEC4COP: Smart Cyber-Grooming Detection Using Natural Language Processing and Convolutional Neural Networks

This paper aims to present the design and implementation of a prototype that recognizes grooming attacks in the context of COP (child online protection) using Natural Language Processing and Machine Learning hybrid model, via Convolutional Neural Networks (CNN). The solution uses a vector representation of words as the semantic model and the implementation of the model was made using TensorFlow, evaluating the classification of grooming for a text (dialogue) prepared asynchronously in a controlled environment according to methodologies, techniques, frameworks and multiple proposed techniques with his development described. The model predicts a high number of false positives, therefore low precision and F-score, but a high 88.4% accuracy and 0.81 AUROC (Area under the Receiver Operating Characteristic).