Research And Implementation Of Personalized Push Service Based On Mobile Terminal

In the big data environment,companies already have a huge retail customer information data and order data.The number of retail outlets has exceeded 8 million nationally and a large number of order information is generated each month.Over time,this order information data volume has reached the level of TB.And a large amount of data to facilitate enterprises to better understand the retail customer details and market dynamics.However,how to apply the existing huge amounts of data is still bothering the enterprises.In the face of massive information data at this stage,the recommendation system appears.The recommendation system is a technique that makes use of historical data for correlation analysis and is a very promising service technology for handling data overload.In recent years,the popularity of mobile terminals represented by smartphones and the number of users have rapidly increased,and mobile terminals have taken an increasingly important role in acquiring information.To solve the problem of information overload,collaborative filtering is the most successful and most widely used personalized recommendation technology in the recommended system at present.And it has achieved rapid development both in theoretical research and in practical practice.However,the traditional collaborative filtering algorithms face many problems,such as the sparseness of the original evaluation data,the difficulty of measuring the similarity between users,and the poor scalability of the system,which affects the recommendation effect.Facing the above problems,the paper proposed the following algorithm to optimize the original filtering algorithm.The main work is as follows:1)Aiming at the difficulty of measuring the similarity between users to the application of traditional algorithms,a measure of the degree of confusion among users is proposed to measure the similarity between users.The first is to calculate the difference between users,that is,the difference between two different user scores of the same project.Second,the difference between the scores is weighted entropy computing to represent the similarity of user ratings.At the same time,pay attention to the impact of active users on the common scoring project in calculating the user similarity,which can reduce the impact of active users on the size of intersection circles as much as possible.Experimental results show that the proposed algorithm can alleviate the problem of inaccurate measure of similarity in the case of sparse data and improve the accuracy and recommendations of similarity between users under the condition that the original score data are sparse and invariable.2)Aiming at the limitation of poor scalability to the application of traditional algorithms,this paper proposes a SVD based k-means collaborative filtering recommendation algorithm.The system only needs to store the singular value matrix of user or item so that the dimension of the eigenvectors of users or projects can be greatly reduced.Then it can ensure the recommendation accuracy and save more storage space.The algorithm takes advantage of the potential relationship between users and projects to overcome the sparsity problem.At the same time,the clustering method retains the advantages of good real-time and scalability.The experimental results show that the proposed algorithm effectively solves the problem of poor scalability of the traditional collaborative filtering recommendation algorithm.3)Setting up a service architecture named Web Service and a system for a variety of mobile terminal platforms.By using the features of the notification bar reminder of various terminal supports,a permanent mechanism of background connection of the client is designed so that the user information message can be pushed according to the real-time scenario even when the client is offline.Finally,the mobile terminal is built to verify the performance of the algorithm and the function of the entire recommended system.