Loading...

Temporal Analysis of Customer Satisfaction based on Steam Game Review Metadata

Seradj, Sepand | 2022

108 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55426 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Heydarnoori, Abbas
  7. Abstract:
  8. The length of time a user plays a computer game is one of the most important metrics for a game’s success in the gaming industry. The importance of this metric can be explained from the viewpoint of different actors present in this industry, for example, from the point of view of the game Developer, increasing the duration of the playtime leads to an increase in opportunity for further in-game marketing, gamers value this metric since it entails how long can a game can entertain them. Making sure that a game can be entertaining for a long period of time is one of the most important factors in choosing to buy a game. Also, it’s in every online store’s best interest to always offer games to potential customers that will entertain them for the longest time. In practice, the duration of the playtime is a of representation of satisfaction of users with a game. Considering the importance of this parameter in the gaming industry, we decided to design a system that can predict how long a given user will play a game that he/she has never played before. Previous Studies conducted on the steam platform have been very few and far in between, often lacking deeper social and technical knowledge with the platform’s user base. In this study, we tried to improve previous research by proposing a new method for predicting the playtime variable based on the data obtained from the social relationships of users on the Steam platform. In this method, in addition to the parameters that have already been used to study this parameter, such as the use of playing time in games with a similar genre or developer, the behavioral information of a user’s friends has been incorporated in such a way that it is measured How effective the playtime of a user's friends can be in predicting the playing time of that user. We trained 8 machine learning models and 3 deep learning models with 65 million data points obtained by preprocessing and engineering the features of a benchmark dataset in this field. Among the machine learning models, the XGboost algorithm with the highest r2 score of 0.78 was the most accurate, and among the deep learning models, a model with a 5-layer perceptron architecture, each containing 16 neurons, was selected with the highest r2 score of 0.75. It can be concluded that despite both models having relative success in predicting playtime, XGboost has performed the best. Also, with statistical studies, we show that the playtime of a user’s friends has very little effect the the user's playtime. We also found out that a user’s playtime among different games from the same developer can be the predictor for other games of the same developer.

  9. Keywords:
  10. Machine Learning ; Deep Learning ; Numerical Prediction ; Steam Platform ; Computer Games ; Playtime

 Digital Object List

 Bookmark

...see more