DETECTING CYBER GROOMING USING TEXT MINING, SUPPORT VECTOR MACHINE, NAÏVE BAYES AND RANDOM FOREST

computer science

The problem of online grooming has emerged as a notable apprehension in present-day society due to the increased use of the internet. This poses a threat to children, as they can be targeted by predators. In order to address this problem, we conducted a research study that utilized text analysis techniques to identify predatory messages. The goal of this research was to protect children from potential harm caused by paedophiles. We aimed to identify specific features and words that are indicative of predatory behaviour, which would enable us to accurately detect such messages in online conversations. By doing so, we aimed to enhance internet security for young children and eliminate grooming incidents. To classify adults who pretend to be children, we focused on identifying crucial features, with foreign words being particularly important. The dataset used in our research was collected from PAN, a well-known source for such data. In order to develop our model, we employed three different algorithms: Naïve Bayes, Random Forest, and Support Vector Machine. Through our findings, we were able to demonstrate that while it is challenging to distinguish between genuine children and adults posing as children within chat logs, we achieved an accuracy of 76.80% in identifying fake children using our best-performing model, SVM. This report discusses the accuracy of the methods we proposed and highlights the essential features that contributed to their success. The primary focus of our study was on detecting grooming conversations, but future research could involve identifying adults who pretend to be children or creating fake profiles. It is important to continue exploring and developing methods to protect children from online grooming and ensure their safety in the digital age.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.