Detecting Cyber-luring Users in the Metaverse Using Non-Text Dataset

Abstract

There has been an increase in the number of cyber-luring victims for child prostitution and child pornography. One of the important social issues is to detect and mitigate these risks as soon as possible. Most of the existing techniques for automatic detection of luring behavior are based on machine learning approaches and have been developed on the premise of using text data such as messages between individuals. However, new problems have arisen in recent years, such as the diversification of contact opportunities in virtual environments and the restriction of the use of conversational corpora due to the growing awareness of privacy protection. In this study, we develop a new method for detecting cyber-luring users using only non-text data such as metadata and user relationship network data. The proposed method is based on graph neural network technology, combines stacking and imbalance learning techniques to capture various contact opportunities among users and detect a small number of cyber-luring users. As a result of an experiment to construct a prediction model for cyber-luring users based on actual data, we succeeded in constructing a relatively high-performance model.