Research And Implementation Of Profiling Algorithms For Mobilr Users On Android System

User profiling technology analyzes and mines user behaviors from a large number of real groups,and builds models to characterize users systematically and completely.It has been widely APPlied in various industries and has become a key area of academic research in recent years.With the rapid development of communication technology,smart phones have become an indispensable device in people's daily life,and the data carried on them directly reflects the user's personal information,such as gender,age,and even personality and hobbies.Therefore,the research of mobile user profiling is more accurate and more valuable.APPlying with machine learning and deep learning technologies,Mobile user profiling algorithms has developed rapidly and made many achievements,but still faces some challenges.First,mobile data security issues.The existing algorithms are mostly collecting data from the phone,and computing on web server at the cloud,and the security risk is likely to occur when the data is acquired and transmissed.Second,the accuracy and sensitivity of the algorithm needs to be improved.The user's dynamic attributes,such as points of interest,living habits,etc.,are instability.How to accurately identify in a short time and flexible adjustment when attributes changed is extremely important.Third,the combination of mobile user profiling and recommendation systems needs to be innovated.The most popular Application scenario of profiles is intelligent recommendation.It is very important to APPly mobile user profiles to help the recommendation system to achieve more accurate and more accurate results.In order to solve problems,which include the difficulty of data acquisition,the safety risk on data transmission and the insufficient of the accuracy and sensitivity of the dynamic attribute detection,this paper proposes a new mobile user profiling algorithm to achieve more accurate and sensitive detection and prediction.We focus on two attributes,wake-up time and sleep time,and propose BTP algorithm to predict the wake-up time and sleep time based on screen state data.It calculates on the mobile phone and data do not upload to the cloud,which avoid the dilemma of collecting data without violating the privacy protection regulations.The experimental results show that BTP can accurately predict the user's wake-up time and sleep time.The MAE of wake-up time is about 10 minutes,and sleeping time is about 20 minutes.Then,for the problems of the lack of accuracy of recommendation system and the problem of serious homogenization of recommend results,this paper proposes a new recommendation method to achieve more accurate and more diversity results.Focus on mobile APP recommendation,we APPly neural network method to learn the characteristics of user features,and propose the P-SNN framework to improve the accuracy of APP recommendation.In addition,we propose the DAM(Dynamtic Adjustment Method),which applies dynamic adjustment to solve the problem of recommendation diversity.Experimental results show that it has lower MAE and MSE,and significantly improves the diversity of recommend results.Finally,we propose an Android APP framework based on the mobile phone user profiling algorithms that consists of four functional modules:data acquisition,storage,calculation,and visualization,which achieves to construct mobile user profiling intelligently,automatically and integrally.