Graph Model Based User Modeling And Recommendation Of Mobile Game

The popularity of mobile devices drives the rapid development of mobile game.Mobile game is becoming an entertainment in daily life of most people.Many mobile game operating platforms have to face the problem that how to mine what users are interested in and what users would pay for as they have a variety of mobile games on them.A personalized recommendation system of mobile game can help them solve the problem.It not only improves the users' feeling,but also helps the operating platforms improve user retention rate and reduce the cost of marketing.User modeling is the core of personalized recommendation of mobile game and depends on the complete characterization of user's interest in mobile game.Most mobile game users have three characteristics: 1)the number of games they played is relatively small;2)the life cycle of their game is longer;3)they prefer to play new mobile game.The characteristics result in the extreme data sparseness of mobile game which brings a great challenge to user modeling and personalized recommendation.This paper models the user's interest in mobile game based on bipartite graph model and use the mobile game records of users to describe the edge weights of the graph.Through the structure of the bipartite graph we can mine the potential interest of the users.On this basis,three recommendation algorithms based on graph are designed and implemented:(1)Preference Based for Bipartite Graph Projection--PBBP.In order to describe the relationship between nodes more accurately,PBBP incorporates users' mobile game preference into the projection process of the bipartite graph to enrich node information of transfer matrix.The algorithm can achieve accurate recommendation in the case of sparse data.(2)Bipartite Graph Based N Projections for Rank--BBNPR.BBNPR is a bipartite graph rank algorithm based on bipartite graph projection algorithm.It can mine the relationship deeply among the nodes so it can locate the user interest precisely and improve the precision of recommendation.(3)Prior Knowledge Based for Bipartite Graph Rank--PKBBR.PKBBR is based on BBNPR.It incorporates users' prior knowledge into the projection procedure of the bipartite graph to enrich the imformation among the nodes.The algorithm can completely describe the relationship of the nodes.Experiments show that PKBBR can further improve the accuracy of recommendation.Based on the graph algorithms,the paper also designs a hybrid recommendation algorithm based on users behaviors.The algorithm distinguishes users based on uesrs' habits and implements different recommendation algorithms for different users.Experiments show that the hybrid recommendation algorithm can further enhance recommendation results.