E-government Deep Recommendation System Based on User Churn

Abstract

At present, all major government-related APPs have basically achieved one-stop operation. However, due to the various categories of government affairs, how to provide users with personalized recommendation services based on user behavior is a problem that smart government needs to solve. Aiming at the problems of sparse user government behavior data and difficulty in mining hidden features, this paper proposes a two-tower model that integrates user churn. A deep neural network is constructed to characterize user item characteristics, and the influence of user churn factor on feature weights is also considered. At the same time, the random forest algorithm is introduced to weight the characteristics of user churn, and the characteristics of the two towers model are combined to achieve personalized ranking recommendation. The experimental results show that our proposed model is better than the original features, and this model has been successfully deployed in the “My Ningxia” government recommendation system, and the user experience has been significantly improved.