Cross-Domain Mobile Game User Interest Modeling And Application

In recent years,mobile game industry is developing rapidly and market competition is increasingly fierce.In this trend,in addition to the quality of the game itself,the game's precision promotion has become increasingly important.The current mobile game marketing platform uses traditional recommendation algorithms to recommend games for those users who have game records.However,when facing users without game records,the platform can only promote games by advertising blindly,which has little effect.This article focuses on cross-domain user interest modeling and try to use the mobile phone users' behavior data in other fields,such as ad click behavior,to portray their interest preferences in the field of game.Then we apply it to potential game users mining and personalized game recommendation.First of all,this paper analyzes the advertising data and game data used in user modeling,and studies the background of data generation and user behavior habits in different fields,which lays the foundation for cross-domain user interest modeling.Secondly,this paper proposes a coarse-grained approach to cross-domain user interest modeling,which is based on domain differences,and describes it by applying it to potential game users mining.At the same time,in order to maximize the utility of user model in potential game users mining,this paper proposes a potential game users mining algorithm named Modify-GBDT(Gradient Boosting Decision Tree)-LR(Logistic Regression),which uses different types of features in different way.Exper iments show that the cross-domain user interest modeling method based on domain difference can effectively extract the user characteristics with distinguishing degree.At the same time,the proposed Modify-GBDT-LR algorithm has better prediction ability than the LR algorithm and GBDT-LR algorithm.Finally,this paper proposes two kinds of fine-grained cross-domain user interest modeling methods: cross-domain user interest modeling based on neighbors,cross-domain user interest modeling based on LFM(Latent Factor Match),and describes it by applying it to personalized game recommendation.At the same time,according to different cross-domain user interest modeling methods,this paper proposes a cross-domain recommendation algorithm based on neighbors and a cross-domain recommendation algorithm based on LFM.Experiments on the actual data set show that the cross-domain user interest modeling method can effectively improve the accuracy of personalized game recommendation,and the two cross-domain game recommendation algorithms can give a better recommendation when compared with popular recommendation algorithm,Item-Based recommendation algorithm and Matrix Factorization-Based recommendation algorithm.