A recommendation system refers to a system recommending or providing information suitable for a user, based on information such as user’s personal information, interest field, and preference. Such a recommendation system applies to various fields such as movies, music, news, books, products, and mobile games.
Representative recommendation systems related with mobile games include Google Play Store, Naver App Store, and One Store. However, such recommendation systems have a problem that actual users’subjective preferences cannot be directly known. To solve this problem, mobile game recommendation systems using Analytic Hierarchy Process (AHP) or Fuzzy-AHP directly reflecting users’subjective preferences were developed. However, the following problems occur:
Concerning an AHP based mobile game recommendation system, a problem that preference level cannot be regarded as the same is caused, although the scale on the items preferred by users is the same. As for a Fuzzy-AHP based mobile game recommendation system, the following problems occur: the mobile game recommendation results preferred by users differ according to membership function of triangular fuzzy numbers, and a suitable mobile game cannot be recommended because AHP hierarchy structure evaluation items (level 2) are not mutually independent.
Therefore, this study solved the membership function’s standard issue by a user’s awarding a probability to the evaluation items of hierarchy structure using a Bayesian network in order to resolve existing mobile game recommendation systems. This study proposed a Fuzzy-AHP based mobile games recommendation system using a Bayesian network not affecting non-correlated items through conditional independence.
The mobile game recommendation system proposed in this study consists of a Web server, database, and client. Four modules are in the Web server of this study, and they are in charge of each process:
The input module inputs the probabilities on the factors through which a user selects a mobile game and inputs the preference for the preferred games. The inference module calculates inference with the Bayesian network using probabilities on the factors through which a user selects a mobile game inputted in the input module. The processing module conducts decision making with Fuzzy-AHP using preference for games preferred by a user, and calculates comprehensive weight using the inference probability of the inference module and relative weight of the processing module. Lastly, the recommendation module calculates priorities using the comprehensive weight calculated in the processing module, and recommends a suitable mobile game to a user by bringing the mobile game information stored in the database.
In the database of this study, probabilities on the factors through which a user selects mobile games inputted in the input module, information on the games preferred by the user in order to compare with recommended results, and mobile game information referring to Google in order to recommend a suitable game from the recommendation module are stored.
The client indicates the main subject using the information provided by the Web server. In this study, the client was structured using mobile application to recommend suitable mobile games to users in real time.
To evaluate the performance of the recommendation system proposed in this study, this study compared it with the already developed AHP based and Fuzzy-AHP based recommendation systems targeting 30 users consisting of college students, company employees, and housewives. As a result, it was ascertained that the recommendation system proposed in this study solved the membership function’s standard issue by users’awarding probabilities to the evaluation items of hierarchy structure. Also, it was confirmed that the recommendation results were more accurate than existing systems by independently revising the hierarchy structure of evaluation items through conditional independence.
In addition, this study comparatively evaluated user satisfaction targeting the same 30-person sample with Google Play Store, Naver App Store, and One Store, which are the recommendation systems currently used by actual users. Consequently, it was identified that the recommendation system proposed in this study showed the highest accuracy and user satisfaction on the recommendation results.