A study of game payment data mining: Predicting high-value users for mmorpgs

Abstract

A user is high-value when his payment amount is more than a predetermined amount (such as 100 dollars). We first analyze payment data under the character level (CLV) dimension of six real mobile MMORPGs and verify the importance of high-value users for achieving game profit and the necessity of HU prediction. Afterward, we propose a CLV-based high-value user prediction (CLVHUP) model that solves limitations of the existing method (delayed identification, the uncertainty of the predetermined amount, and the inflexible time window). This model not only can predict high-value users with sparse features and a small volume of users but is also sensitive to the information underlying CLVs. To the best of our knowledge, this is the first work to analyze payment behavior under the CLV dimension and to predict high-value users by utilizing machine learning techniques and exploring the information underlying CLV. Comprehensive experimental results demonstrate that the performance of our solution is better than that of several baselines.