Multimodal Deep Learning Architecture for Identifying Victims of Online Death Games

Online death games are a fairly recent public health concern of the modern technology-driven world. Various dangerous online games like Blue Whale Challenge and MOMO challenge have grown popular through social networking sites where players or victims engage in self-harming activities, often leading to death. This problem domain has not been studied in depth till date and no known technology-based solutions exist to prevent the spread of such dangerous challenges. The prime objective of our research is to explore the use of deep learning and transfer learning techniques for content analysis of user-generated posts over various social networking sites and design an early warning system which can be used by healthcare authorities for timely identification of victims of these games so as to avoid any fatalities. In this paper, we first discuss in detail the numerous challenges in building required technology-driven solutions for this domain. Next we propose a multimodal deep learning-based system for identifying victims of online death games, using state-of-the-art feature generation techniques for two modalities in user’s social media posts: image and text. To the best of our knowledge, our proposed system is the first technology-driven public healthcare administration tool for this this domain.