Research On Distributed Recommendation Technology For Game Mall

With the development of Internet,online games have become an important part of people's daily entertainment.There are thousands of items in the game mall,so it is difficult for players to find the items they want quickly.Therefore,personalized recommendation system is becoming an indispensable part of online games.Compared with SVD matrix dimensionality reduction algorithm,hybrid collaborative filtering algorithm can effectively avoid core data loss when solving the problem of sparse characteristic matrix,but this algorithm will cost a lot of computing time,resulting in a heavy delay of personalized recommendation,seriously affecting the user experience.In this paper,by using distributed technology to reduce the computing time of hybrid collaborative filtering algorithm,the problems of current mainstream distributed algorithms are analyzed in detail.For example,the consistent hash algorithm cannot guarantee the uniformity of task allocation;The uniform polling algorithm cannot allocate tasks dynamically and reasonably according to the difference of computing capability between servers and the difference of complexity between tasks;The task assignment algorithm,which assigns tasks according to the variability of the server's computing capability,cannot dynamically and rationally assign tasks according to the variability of complexity between tasks.In order to solve the above problems,This paper proposes a load balancing strategy based on dynamic adjustment of task completion.The number of tasks completed in this round of services is monitored regularly,and it is taken as the base of tasks assigned to the next round of services.This load balancing strategy can effectively solve the problem of unreasonable task allocation caused by the difference of computing capability between servers and the difference of complexity between tasks,thus greatly reducing the overall computing time of the algorithm.This paper establishes an algorithm model based on the data characteristics of online games,and divides the behavior of players into the activity behavior and purchase behavior,We use user similarity-based collaborative filtering recommendation algorithms to solve the problem of sparse characteristic matrix of purchase behavior,Collaborative filtering calculation based on items is carried out to get the relationship between items.The algorithm in this paper shows good scalability in the evaluation.When the physical machine reaches 8,the computing time is reduced from 24 hours and 25 minutes to 3 hours and 47 minutes.Compared with the three distributed algorithms such as consistent hashing with bounded loads,uniform polling,and server computing capability classification,the efficiency of the algorithm in this paper is improved by 18%,14.5% and 7% respectively.In addition,in the test environment,compared with the collaborative filtering algorithm of SVD and Borda ranking,the accuracy of this algorithm is improved by 7%.