The accuracy of two recommendation algorithm are 0.67 and 0.86. High accuracy was based on MapReduce and association rules mining of library personalized recommendation method, the method for advanced users, need a more accurate recommendation. Low accuracy was based on MapReduce and user model clustering of library personalized recommendation method, the method for the general customer, when recommended considers more of the efficiency of the library. The experimental results show that both of the recommended method is suitable for the user groups, based on the user recommend a layered strategy fully consider the characteristics of two different customers. Better library services for different users.