Research And Implementation Of User Profile Algorithm Based On Mobile APP

With the rapid popularization of mobile phones,mobile APPs begin to take a part of daily life of users.These apps can collect a lot of valuable terminal data by using a variety of sensors on mobile phones.At the same time,users' demand and hobbies can be analyzed by modeling the functions of app which they use.How to make use of these data effectively to assist business decision-making and advertising has become research hotspots.User profile has always been a powerful tool to analyze the characteristics of users.User profile algorithm can get the user's private attributes and tags by mining user's characteristics and behavior data,which can describe user in the downstream tasks directly.Therefore,how to use a variety of app and user behavior data is a very popular research field for obtaining user profiles.To solve the problem,both academia and industry have proposed a series of methods to model user profile of mobile app.However,current research is still facing many challenges.Mobile users' data generation sources are scattered and diversified.In the mean time,user profile algorithms need to overcome privacy protection issues.First of all,the paper proposes an algorithm that uses users' comments on app to model and analyze user profile.Users pay more and more attention to maintaining their privacy so that it is difficult to obtain user profiles by questionnaire survey and other direct methods.As mentioned above,we can analyze data that user download and comment app to model their interests and hobbies.However,behavioral data of user's operation on mobile phone is user's personal privacy.It's difficult to obtain these data.Fortunately,users' download and comment behavior on app can also reflect user's demand for mobile app.In this paper,we use these two behaviors to design different pre-training tasks,and use pre-training model to get user profile.The pre-training model utilizes user behavior and app function to model user profile.Because user comments and app information contain a lot of text features,this paper uses BERT,which is the most popular model in NLP,to preprocess these features.In paper,user profie is applied to downstream recommendation system,which proves that our algorithm can represent the user effectively.Then,aiming at the problem that computing resources of mobile devices are still insufficient,this paper proposes an algorithm to mine user location data and screen state change data to get periodicity of user profile.The algorithm can be used to calculate location of user's residence,work place and user's schedule.Short-term user profile is calculated by using collected data on mobile terminal device,and then the long-term user profile is calculated by using the short-term user profile on the cloud server.The algorithm is divieded into two parts in logic and architecture.This kind of design reduces computing resources and storage resources needed by algorithm on the premise of ensuring computing effect.It also avoids the risk of violating users' privacy.Finally,app store system based on user profile algorithm is implemented.Users can use app store to browse,manage and download app.The system can calculate user profile and apply it to recommendation system,so as to realize personalized recommendation of app.