Research On Hierarchical Recommendation Of Personalized App Based On Deep Learning

The development of Internet has greatly improved people's ways of life.Internet has become more and more close to people's life.With the decrease of mobile terminal price and the popularization of wireless network,the number of intelligent mobile devices is increasing.It is very important to provide the most personalized app for intelligent mobile users.At present,there are many researches on app recommendation,but there are still three main problems:(1)it is mainly based on the basic information of users and app,and it does not process the dynamic behavior information between users and app,and it does not capture the personalized preferences of users.(2)The app recommended to users is based on the data searched and used by users in the application market before.The association between users and app in this part of data is not found out,which can not meet the needs of personalized recommendation.(3)It mainly uses tree model or neural network structure for processing.Tree model can't learn discrete features well,and neural network structure can't learn continuous features well.In view of the above problems,this paper proposes a research method of personalized app recommendation based on deep learning,with the main contributions as follows:(1)The characteristic analysis of user's dynamic behavior.When the user uses the app,there will be corresponding operation behavior information.This part of information is in a changing process,which is dynamic data information.Another part of behavior information is the basic data of users,which is static data information.From the dynamic behavior information of users,we can capture the changes of personal preferences when users use the app.At the same time,we use the static data information of users as a prior condition to capture the dynamic behavior information of users for correction and improve the overall recommendation accuracy.(2)This paper proposes a DHM model based on the dynamic behavior of users.When users use devices and apps,they have their corresponding operating habits.The dynamic behavior changes of users can reflect the changes of users' personalized preferences.Therefore,we propose a method to capture the dynamic behavior information of users,and process it in combination with static data information.This paper proposes a method combining deep learning model and tree model to recommend.The tree model mainly deals with static numerical data,and the neural network structure in deep learning deals with the behavior information of user type data.At the same time,the batch structure of deep learning can realize the purpose of dynamic updating recommendation effect,and online recommendation can be realized later.(3)A personalized app recommendation model(UHAM)based on user hierarchy is proposed.On the basis of user age stratification,users and app are stratified.User stratification can like users' personalized preferences.App stratification can refine the categories of app and combine the hierarchical relationship between users and app to improve the effect of recommendation.This paper starts from the relationship between user hierarchy and the corresponding level of app,and at the same time,the app is also layered,mining the relationship between user and app levels,and combining the dynamic behavior information of users.By building preference as a measure of tag advancement,the effect of personalized app recommendation model based on user hierarchy is verified.