Search

Online gaming no longer has limited access, as it has become available to a high percentage of children in recent years. Consequently, children are exposed to multifaceted threats, such as cyberbullying, grooming, and sexting. The online gaming industry is taking concerted measures to create a safe environment for children to play and interact with, such efforts remain inadequate and fragmented. Different approaches utilizing machine learning (ML) techniques to detect child predatory behavior have been designed to provide potential detection and protection in this context. After analyzing the available AI tools and solutions it was observed that the available solutions are limited to the identification of predatory behavior in chat logs which is not enough to avert the multifaceted threats. In this thesis, we developed a chatbot Protectbot to interact with the suspect on the gaming platform. Protectbot leveraged the dialogue generative pre-trained transformer (DialoGPT) model which is based on Generative Pre-trained Transformer 2 (GPT-2). To analyze the suspect's behavior, we developed a text classifier based on natural language processing that can classify the chats as predatory and non-predatory. The developed classifier is trained and tested on Pan 12 dataset. To convert the text into numerical vectors we utilized fastText. The best results are obtained by using non-linear SVM on sentence vectors obtained from fastText. We got a recall of 0.99 and an F_0.5-score of 0.99 which is better than the state-of-the-art methods. We also built a new dataset containing 71 predatory full chats retrieved from Perverted Justice. Using sentence vectors generated by fastText and KNN classifier, 66 chats out of 71 were correctly classified as predatory chats.