Research On Recommendation Algorithm Based On Bipartite Network And Implementation Of System

With the development of the Internet of Things,Cloud Computing,Big Data,Artificial Intelligence,and other digital technologies,the world has entered the era of the digital economy.How to help users find the information they are interested in from the mass of information has become a widespread concern in academia.As an effective information filtering system,the recommendation system is an important tool to solve this problem However,with the rapid growth of users and items,the traditional recommendation system based on collaborative filtering is facing the challenge of decreasing recommendation quality and real-time performance.Recent studies have showed that compared with traditional collaborative filtering algorithms designed data feature manually,the network representation learning method can enhance the feature representation ability of recommendation systems.In order to improve the recommendation quality and real-time performance of the recommendation system,this thesis focuses on the recommendation algorithm based on the bipartite network and the architecture of real-time recommendation system.The main work and contribution of this thesis are as follows:(l)A bipartite network representation learning regularized matrix factorization recommendation algorithm(BiNRMF)is proposed.Bipartite network representation learning is used to learn the low-dimensional dense representation of users and items to improve the stability of similarity calculation results and reduce the complexity of similarity calculation;the similarity between users and items in the low-dimensional vector space is integrated into the traditional matrix factorization algorithm to train the prediction model.(2)A real-time three-layer recommendation architecture is designed.The overall architecture is divided into three layers:offline layer,near-line layer,and online layer.The task of the offline layer is to train the recall-model and calculate the recall results.The near-line layer uses real-time data to update the parameters of the rank-model.The online layer provides users with real-time services and generates the final recommendation results The open-source Wide&deep model in Tensorflow is used to improve the training efficiency and service response efficiency.A real-time recommendation system application that includes the client-side and administrator-side is implemented with the Spring Boot framework and front-end technology.