Prediction of Churning Game Users Based on Social Activity and Churn Graph Neural Networks

This study introduces an innovative churn prediction model that leverages player activity and social interaction data from the massive multiplayer online role-playing game (MMORPG), Blade and Soul. This model uniquely visualizes player interactions as a graph structure and enhances prediction accuracy by integrating a graph convolution network (GCN) and correct and smooth (C&S) techniques into social network analysis. Focusing on the intrinsic features within the graph structure, the GCN delves into internal dynamics, whereas C&S synergistically incorporates external label propagation, considering the influence of neighboring players. The amalgamation of these methodologies increases the precision of churn predictions by considering both internal user characteristics and external social influences. This finding highlights the critical role of social activities in understanding player retention in MMORPGs. This study contributes significantly to the gaming industry by demonstrating how integrating social data into churn predictions can aid in the early detection of player disengagement, thereby bolstering the sustainability and growth of gaming services. Furthermore, it offers a novel framework for comprehensively comprehending and analyzing player churn by applying social network analysis and graph theory, providing profound insights into this complex phenomenon.