Research On Personalized And Accurate Delivery Algorithm Of Email Advertisement Based On Deep Learning

With the rapid development of Internet and the wide coverage of mobile Internet,internet has become one of the most important channels for the public to access all kinds of information.In recent years,a large number of internet companies have launched various personalized recommendation services.Advertisement,as the main income of internet companies,is without exception.Many companies have done a lot of experimenting of personalized recommendation of advertisement business.However,the current research on personalized advertisement recommendation is mostly in the context of information flow.There is little research on the scenario of email,compared with others.Email is the first and most widely used internet application of all internet applications.Every user of the internet has at least one e-mail address,some have even more.Therefore,the research on personalized and accurate delivery of email advertisement is very necessary and has its practical value.The research content of this thesis is based on the mailbox platform,combined with the current popular technology—depth learning.This thesis studied the personalized and email advertisement circumstantially and implemented a system of email advertisement's accurate delivery.Also,the thesis put forward a workable idea and promotion algorithm for the personalized recommendation of advertisement business.The main work of this thesis is as follows:(1)A comprehensive user interest profile is established.In addition to the two regular characteristics which are users' personal information and users' behavior information,this thesis proposed to use LDA model to extract the subject of users' personal e-mail content and then add these subjects as interests to user's interest profile.This process complemented the shortage of email users' interest source.(2)A UIP-DeepFM algorithm based on user interest profile is proposed.The algorithm is based on the varied user interest profile established above.It used the parallel structure of factorization machine and deep neural network.Both parts shared the only input,which made the algorithm more efficiently.The part of factorization machine is responsible for the effective combination and learning of the first-order and second-order features.The other part of deep neural network is responsible for the deep mining and extraction of the high-order features.Finally,the algorithm output the predicted click through rate.(3)An intelligent precise delivery system of email advertisement is built,which included three modules:sending rate adjustment module,delivery volume of provinces adjustment module and target user selection module.This thesis has built a data set for experimental test due to the few open data set of email advertisement recommendation.The dataset used in this thesis came from the real user data of a mailbox company in China.After pre-processing,such as filtering and cleaning,the training set and test set were obtained.Finally,the experimental results showed that the system had a very good performance and application value for email advertising recommendation.