Research On Direct Mail Precise Delivery Model In Cloud Computing

In recent years,with the maturity and development of high-techs such as cloud computing,big data and Internet of Things,traditional direct mail(DM)advertising media industry has been transformed rapidly.In the advertising model of cloud environment,DM advertisement and user data volume has increased dramatically.This can cause data sparsity and algorithm scalability problem,which has become an urgent challenge to be solved.Traditional advertising is currently building the model from two aspects,user behavior and content orientation.Dividing users by similarity calculation,and delivering the advertisement according to user common behavior or preference content.The extreme sparsity of data sets in the cloud environment causes the rarity of effective data for model training,so we can't find the nearest neighbor of users or advertising exactly.Meanwhile,the calculation of large amounts of data needs a lot of computing resources and time,causing the poor scalability of the traditional method.Aiming at drawbacks of the traditional precise delivery model,this paper mainly focuses on DM advertising in the cloud environment.Combing with the bisecting kmeans and L-BFGS optimization of collaborative filtering algorithm,the DM precision advertising model is proposed in this paper.The core structure of the paper is as follows:(1)Focusing on the DM advertising process,the paper analyzes the differences between the traditional environment and cloud environment advertising process.According to the methods and difficulties of accurate delivery under the cloud environment,the key problems are proposed in this paper.Then the two-stage algorithm of cloud environment precise delivery model is designed.(2)According to real-time cluster analysis of user characteristic and behavior,analyzing the conversion and clustering of existing data sets is set as the data input of collaborative filtering algorithm in the precise delivery model,studying the data processing flow of precision advertising model and basing on bisecting k-means and Hadoop distributed data warehouse.(3)The paper proposes the L-BFGS optimized collaborative filtering algorithm as the prediction method of the precise delivery model,combining the feature of users and commodities,using the factorization machine model(FM model)to predict the extent of users' preference for advertising and reduce the data sparsity effect at the same time Using L-BFGS algorithm as the parameter training method of FM the model,which has fast convergence and low space occupation of calculation,and is able to improve the training FM model computing efficiency and scalability to meet requirements of prediction accuracy and computational efficiency in the cloud environment.Finally,basing on the Spark,precise delivery model is achieved and numerical experiments is designed for verifying the model.By comparing the experimental results with different data sets and the other three methods,and analyzing the effects of the convergence speed,the number of features and the size of the data set on the accuracy of the algorithm,it could be concluded that the model in all the four different scale data sets have the highest prediction accuracy,and better results could be achieved by an appropriate increase in the number of iterations and the characteristic of data sets,which verifies the validity of precise delivery model.This paper proposes a new way of thinking of the problem of the precise placement of DM ads under the condition of extreme sparsity of data in cloud environment.