International Conference on Engineering, Applied Sciences, and Technology August 21‐24, 2013, The Sukosol, Bangkok, Thailand Online Game Player’s Behavior by using Hidden Markov Model Kittipat Savetratanakaree a , Kingkarn Sookhanaphibarn b , Sarun Intakosum a a Department of Computer Science, Faculty of Science, King Mongkut’s Institute of Technology, Ladkrabang, Thailand. b Department of Computer Science and Software Engineering, School of Science and Technology, Bangkok University, Thailand. Abstract—Most game operators revenues come from the subscription fees and sale of virtual items. Game operators need to understand the player’s behavior and prediction of player’s departure. This paper presents Hidden Markov Models (HMM) approach for pattern recognition of online game player’s behavior in game revisitation, classified into player’s behavioral state. HMMs have been used for pattern recognition and classification problems since HMMs are proven suitable for modeling dynamic systems. Online Games are dynamic systems which games can be changed according to players can play games with selected different characteristics or roles in game (role-playing game). A large-scale analysis to find player behavior in game revisitations is conducted in our research by using Shen Zhou Online access log. These logs were collected for nearly 4 years consisting of 50,000 player accounts. We use the information on game revisitations, together with daily playing time such the login time, staying time and login frequency information. In this paper, we focus on the predictable records on the past data set of player’s behavior in revisitation to the game using HMMs into behavioral states. Keywords: HMM, Revisitation, Player Behavior, Online game, Shen Zhou Online I. INTRODUCTION Massively Multiplayer Online Role-Playing Game (MMORPG) has become increasingly popular in recent years. Shen Zhou Online (SZO) in Taiwan is categorized as MMORPG. SZO is developed and distributed by UserJoy Technology Co.Ltd., SZO maintains at any moment thousands of online players, who must purchase “game points” if they wish to continue their game adventures in the virtual world beyond the 30-day free trial period. Game industry’s revenues mostly come from the subscription fee and virtual items sale. The game operator prefer to have more loyal “hardcore players” who would stay in a game more than a year rather than players temporarily or permanently absent from the game after the end of 30-day free trial period. The Online game player’s behavior model [1] is needed for prediction when player’s departure from the game. The game operator might give any special promotions to those players who are expected to be absent from the game in order to continue playing the online game which increases the player retention rate of the game. This analysis is important to further study in prediction when the player will absent from the game. This paper objectives are 1) Understanding the player’s behavior in observed player’s information on game revisitation 2) Developing a set of states presenting online game player’s behavior by using Hidden Markov Model According to the previous studies [1, 2, 3, 4, 5], game revisitation is the situation that game players returns to play the game again after they might stop playing game for some periods of time. They might be busy or they might do some other things more interested during the stopped periods. It is interesting to know that how long they will return to play the game or they might quit the game. There are typical patterns exist in online-game players’ revisiting to a game of interest and its areas. The Hidden Markov Models(HMM) were first introduced in 1970 as a tool in speech recognition which has become increasingly popular in the recent several years because of its strong statistical foundation for use in a wide range of application as in pattern recognition such as speech signal recognition [7,8,9], handwriting recognition [10], gesture recognition, stock market forcasting [11]. A HMM can be considered the dynamic Bayesian network which provides a probabilities framework for modeling a time series of multivariate observations. We applied HMM with online game player’s information on game revisitation because players can change their role-playing as different characteristics in each login which HMM has ability to handle new data robustly and efficient to develop observed sequences to hidden states we are considering. In this paper, we explain our definition of HMM in Section II. We focus on predictable dataset of online game players records. We filtered unpreferable records out of the predictable dataset and use HMM based model with observed sequences to develop a set of states of game player’s behaviors in revisitation to the game displayed as a finite state machine diagram in Section III. We will extend our approach for further study to predict the player’s departure from the game which will be our future research. II. HIDDEN MARKOV MODEL A Hidden Markov Model (HMM) [6] is a Markov Model in Statistics that the system being observed or modeled assumed as a Markov process. There are observed sequences that can be evaluated as hidden states. These states 274