Research On Telecom Package Recommendation Model Based On Deep Learning

Throughout the world,with the gradual marketization of telecommunications industries in various countries of the world,competition in the telecommunications enterprices has become increasingly fierce.The official implementation of the number portability has intensified the battle for users among the three major domestic telecom operators.In order to expand their respective market shares and enhance their core competitiveness,operators have formulated various telecom combination packages in anticipation of being able to solve the differentiated needs of users at different levels.It is a very important research topic to study the user's package selection behavior in the field of telecommunications marketing.How to provide users with better services according to their own actual needs,can firmly grasp the existing users,and improve users' satisfaction with products Degree and long-term support.However,although a wide variety of packages has brought more choices for users,at the same time it will also cause some problems: when new or old users choose or change packages,facing so many telecom packages,they are at a loss to choose the right one in a short time.Therefore,it is particularly important for telecom companies and personal user experience to accurately and intelligently recommend packages for users.In recent years,with the rapid improvement of computer hardware equipment and the accumulation of massive data,the research on deep learning theory and technology has become more and more popular.For the recommendation of telecom packages,the traditional recommendation method either requires complex artificial feature engineering,or cannot use deep learning's powerful feature extraction capabilities for end-to-end deep learning training.This paper utilize deep learning models(especially the Wide & Deep models in the field of Click-Through Rate Prediction)to implement telecom package recommendations,and has achieved good results in both evaluation standards AUC and Log Loss.In addition,based on the deep and cross network and the deep factorization machine model,this paper proposes two recommendation models for telecommunication packages that can process text information.Both models are end-toend deep learning models and avoid artificial feature engineering.In these two models,we have replaced the feedforward neural network of the deep component with the deep belief network,and have integrated the Word2 vec model.This makes the improved two models not only effectively capture cross features and deep hidden high-order features,but also make full use of the text information in the data to further improve the model's recommendation ability.This paper has been experimentally verified in the real data set provided by China unicom research institute.Compared with several mainstream recommendation algorithms,it has better recommendation performance.