| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系 === 106 === We want to analyze the players interest on a gamble game by studying their behavior, that is, when will they stop playing a game. We checked several common distributions. All of them were not fit on players’ total play time, because they cannot pass the Kolmogorov-Smirnov Test. However, we discover that the logarithm of the data is suitable for linear regression. In fact, the correlation coefficient is very well on all games. That is why we deicide using it for the follow-up analysis.
After linear regression, every game will have two parameters, slope and intercept. We picked the slope of regression line as the characteristic value of the game. With the player’s playing time, we can get the score which means a player’s preference of the game. The player scores a game by how long he played. Now, we can build a recommend model by collaborative filtering with the scores.
We try to deal the cold-start problem which is often encountered on recommend system. We estimate a player’s total game time after he played a relatively short time. And compare the score between real playing time and the estimate time from the recommend model. We found out the error is very small. This is because the basis of the recommend model is based on the player''s playing time.
There are two contributions to this dissertation. First, we propose a method to convert user behaviors into feature values to quantify “preferences.”. Second, there is not much research on statistical analysis of user behaviors of gamble games in Taiwan. The ideas from this research can be used as a reference for other studies.
|
|---|
