Research Of Personalized Recommendation Methods

Recommendation system is a powerful tool to overcome the problem of filtering from too much information.We build recommendation system trying to recommend items or services that users require.Considering the actual scene of recommender,We have proposed two different personalized recommendation methods in term of different views.One method is called user profile hybrid recommender which builds user profile with the help of taking all kinds of users' data into consideration.In order to mining the connection of users and items from users' behavior data,translation based preference scaling model has been proposed.The core of this paper is translation based preference scaling recommendation model which is a recommender method based on presentation learning.Inspired by the translation model in knowledge bases,we model the translation between the users and items.This model aims at learning distribution presentation of users,items and user behaviors.Besides,we not only regard user behaviors as different vectors,but also try to scale the preference of different user behaviors with their actual features.Therefore,translation based preference scaling recommendation model can be applied to both explicit and implicit feedback which overcoming the shortcoming of matrix factorization recommender method.Experiments based on different dataset have turned out that the proposed methods are feasible and effective.