Design And Implementation Of Product Recommend System Based On Deep Learning

With people's increasing attention to the user experience of online shopping,the two major issues of product recommendation accuracy and system concurrency have become the research focus of system developers.At present,the mainstream product recommendation algorithm is a content-based collaborative filtering algorithm,but this algorithm is insufficient in the field of mining data features,and a single algorithm cannot effectively express all the user's information.Similarly,the system concurrency is also a system performance problem that needs to be improved.Aiming at these two problems,the product recommendation system mainly researches the optimization of the recommendation algorithm and the optimization of the system performance.At the same time,it implements the basic functions of online shopping to facilitate the system to collect user data.The product recommendation system develops system functions based on the microservice architecture,and completes recommendation-related services,user-related services,product-related services,and order-related services,ensuring low coupling,high cohesion,and strong expandability of system functions.Among them,recommendation related services include personalized recommendations,related product recommendations,and popular product recommendations,broadening the breadth of recommended content,diversifying the information that users feedback back to the system,and reducing the fitting of personalized recommendation models to a certain extent.The recommendation model improves the previous content-based collaborative filtering algorithm model.It uses a probability matrix decomposition model(model 1)fused with deep learning and a user-based collaborative filtering model(model 2).Model 1 improves the content-based coordinated filtering algorithm.In the field of feature extraction,two deep learning models,Attention-CNN and LSTM,are used to obtain the prior probability distributions of the user and product implicit feature vectors,so that the training data better reflects the data features and the training results are more accurate.User historical data mining users 'past preferences.Model 2 is a user-based collaborative filtering algorithm.Based on the data of similar users' potential preferences,the fusion of the calculation results of the two models can provide more accurate and reasonable personalized recommendations for users.To improve system performance,The whole cluster system deployment,service and database design before the cache architecture,a shared data reading and writing pressure,speed up the response to the request,and deploy a Hadoop cluster,the recommended candidate set for off-line calculations,reduces the effect of external services.The implementation and testing of the system show that the basic functions of the product recommendation system can work normally,and the personalized recommendation algorithm used in it is more accurate than other personalized recommendation algorithms.The architecture design of the system cluster deployment and cache can be obvious improve system performance.