Analyzing User Behavior Patterns in Casual Games Using Time Series Clustering

Abstract

User behavior pattern classification based on time series clustering is playing an increasingly important role in the game industry. However, user behavior and data performance are quite different in lightweight casual games, compared with those in MMORPGs (Massively Multiplayer Online Role-Playing Games) that many research institutes studied before. With the development of mobile devices and the fragmentation of users' time, both the number of users and the importance in today's gaming industry for casual games jump rapidly. The unique user data performance, such as high sparsity, poses new challenges to clustering time-series data of user behavior based on this kind of game. In this paper, we take UNO!, a mobile card game with hundreds of millions of users, as our research object, and propose an improved time series similarity measurement via the smoothed sequence Euclidean distance to realize clustering analysis of user behavior patterns. In this analysis, we purposefully use the user feature sequence that is more consistent with the game feature. Finally, we explore the correlation of clustering results with user payment, and propose a visualization scheme that can comprehensively show users' payment behavior, including short-term and long-term, and its relationship with users' game behavior.