Design And Implementation Of APP Recommendation System Based On User Behavior

With the rapid development of mobile computing,in order to facilitate users to use smart mobile devices and access information services in the mobile network,the number of apps has increased dramatically.Such a large number of applications have made it impossible for users to quickly select apps that meet their needs and meet their potential interests from these massive applications.In order to solve this problem,each application market applies the personalized recommendation system to its own application market,and actively recommends apps that meet users' potential interests.Therefore,app recommendation system has become a hot issue that is widely concerned by the current research.Collaborative filtering is the main recommendation algorithm of app recommendation system,but the traditional collaborative filtering algorithm has the problem of sparse grading data.Aiming at this problem,this subject designed and implemented an app recommendation system based on user behavior,and its main work is as follows:(1)The collaborative filtering algorithm based on scores is studied,mainly including the collaborative filtering algorithm based on memory and the collaborative filtering algorithm based on model.Collaborative filtering based on user,item and regularized singular value decomposition(RSVD)is implemented.Combined with experimental data,the three algorithms were evaluated from the four evaluation indexes of accuracy rate,recall rate,F1 value and SSD.(2)Aiming at the problem of sparse grading data,this thesis improves the collaborative filtering based on grading.A collaborative filtering method based on user behavior sequence is proposed,which takes richer behavior log data as the input data of collaborative filtering algorithm.The user behavior sequence is constructed by data preprocessing and the similarity calculation method based on the behavior sequence is proposed.Combined with experimental data,the improved algorithm was compared with the three collaborative filtering algorithms based on scores in the four evaluation indexes of accuracy,recall rate,F1 value and SSD.(3)To solve the problem that users' interests will change with the recommendation of time,this subject further improves the collaborative filtering based on user behavior sequence,and proposes a collaborative filtering based on user behavior sequence and time attenuation.Considering the different influences of different time data on users,this thesis proposes a calculation method for each behavior sequence based on time weight,and compares the algorithm with the collaborative filtering based on user behavior sequence on the four evaluation indexes of accuracy,recall rate,F1 value and SSD.(4)Design and implement an app recommendation system based on user behavior,including data acquisition module,data preprocessing module and app recommendation module.In the process of data collection,data crawling on mobile terminals is involved,while in the app recommendation module,a collaborative filtering algorithm based on user behavior and time attenuation is used to recommend apps for users.Through the initial use of this system,the time span for the 2017-08-08-2017-08-08 user behavior data and app text data as the input data system,choose one of the 1000 users and 5281 app interaction with the user(the user rating data sparse,and behavior of log data is rich),recommend app,for the 1000 users recommend the results meet the sparse data of users for scoring personalized recommendation effect.