Research And Application Of User Profile In Content Push

In the era of big data,it is particularly difficult for mobile users to find content services that are interested in them from the mass of information,and how the content service providers can target the user community and serve the user better.To solve this problem,this topic achieve a subsystem that can automatically provide personalized content services.The first step is to collect the data generated by the mobile user,that is,the mobile data,and divide it into static data and dynamic data according to whether the data is change or not.The static data is the basic information of the user.The dynamic data is user behavior data:interest data,mobile App data,location data,the using of intelligent terminal data,and according to different date types,to build a tree structure of the taglibrary.The background system then forms the content library by editing the different content and then organizing the content into meaningful content services and mapping it to the corresponding tag.On the basis of the labeling system and content library,and according to the the law of daily activities,the day will be divided into eight different time periods,such as working hours,lunch time,rest time,and then statistics the number of interest tags in every periods,and for different data types using different calculation methods.Interest data,using custom formula calculation;mobile application App data,using the improved TF-IDF(term frequency-inverse document frequency)algorithm calculation;location data and the data of using intelligent terminal using statistical methods;the value calculated is viewed as the weight,the bigger the value,the greater the degree of preference for the label,and then sorted,select the Top-N labels,as the user's individual profile;Based on the results of user profile,the classification algorithm is used to predict the interest of different sex and different age users in different time situations.This paper studies the use of KNN(K-Nearest),SVM(Support Vector Machine),BP(Backpropagation)neural network,DNN(Deep Natural Network),and experiment on the Iris dataset and mobile data,by comparing the accuracy and time-consuming of algorithm,finally,the DNN algorithm is selected as the prediction algorithm.Finally,combined with the user's current location and time scenarios,through the corresponding push algorithm,the use of location scenarios,time scenarios followed by the strategy,the use of user profile and predictions of interest-loving labels,select the contents from the content library,automatic push to the user.Then it is proved that DNN-based personalized push subsystem can provide personalized content push service based on user's position change and time scene change,and has better system performance compared with traditional push service.