User segmentation for retention management in online social games

Abstract

This work proposes an innovative model for segmenting online players based on data related to their in-game behaviours to support player retention management. This kind of analysis is helpful to explore the potential reasons behind why players leave the game, analyse retention trends, design customised strategies for different player segments, and then boost the overall retention rate. In particular, a new similarity metric which is driven by players' stickiness to the game is developed to cluster players. Three feature dimensions, namely engagement features (e.g., log-in frequency and length of log-in time), performance features (e.g., level, the number of completed tasks, coins and achievements), and social interactions features (e.g., the number of in-game friends, whether or not to join a guild, and the guild role), are employed and aggregated to derive the stickiness metric. The applicability and utility of this new segmentation model are illustrated through experiments that are conducted on a realistic MMORPG dataset. The derived results are also discussed and compared against two benchmark models. The results reveal that the new segmentation model not only achieves better clustering performance, but also improves player's lifetime prediction by better distinguishing between loyal customers and churners. The empirical results confirm the effects of social interaction, which is usually underestimated in the current research, on player segmentation. From an operational perspective, the derived results would help game developers better understand the different retention-behaviour patterns of players, establish effective and customised tactics to retain more players, and boost product revenue. � 2017