Design And Implementation Of A Fresh Food E-Commerce Platform Based On Micro-Service Architecture

In recent years,with the change of user consumption concept and the development of cold chain technology,the fresh food e-commerce platform has become an important fresh commodities marketing channel.However,due to the expansion of business and the surge of users,the fresh food e-commerce platform is facing many challenges.First,the traditional fresh food e-commerce platform has poor performance and poor system stability,which cannot cope with high concurrency scenarios.Second,in the face of users’ demand for timely acquisition of fresh commodities,the platform lacks the ability to locate stores around users with high performance and remind stores of real-time replenishment.Thirdly,to create high-quality fresh food purchase service,the platform lacks high-precision personalized recommendation ability suitable for this business scenario.To solve this problem,combined with personalized recommendation algorithm,the thesis designed and implemented a fresh food e-commerce platform based on microservice.Fresh food e-commerce platform mainly includes store service,recommendation service,category management service and commodity service.Among them,the store service is responsible for efficiently searching for stores around the user,reminding the replenishment.Specifically,to improve the retrieval efficiency of stores near the user,the store service has designed a high-performance retrieval solution for surrounding stores based on the Geo Hash algorithm.The category management service is responsible for the efficient management of category information.The category service builds distributed transaction coordinator based on Seata to ensure data consistency and realize system stability under the microservice architecture.The commodity service is responsible for efficient management of commodity information to improve the overall function of the system.The recommendation service is responsible for building high-precision personalized recommendation services for the platform.In order to effectively help the recommendation model get the best effect,the service uses Spark,Kafka to complete realtime and efficient data collection,and uses clustering algorithm to mine users’ high-order hidden features.In the recall layer,this thesis uses Youtube DNN model as the basic model,and proposes a commodity recall model based on long-term and short-term interests.The model introduces long-term interest sequence features to enrich the expression of user interests and it combines the Temporal Convolution Network(TCN)to mine the temporal evolution process of users’ short-term interests.At the same time,the model proposes seasonal sensitivity factors and purchase cycles to weight the similarity calculation rules.In the fine sorting layer,the thesis improves the AFM(Attentional Factorization Machine,AFM)model by adding an automatic discretization structure.In the offline training experiment based on the internal data set of the internship enterprise,compared with the basic model,the Hit Rate@100 index of the fresh product recall model based on long-term and short-term interest improved by 2.23%,and the AUC of the AFM model based on the automatic discretization of numerical features improved by 2.79%.After the implementation of the system,this thesis conducts functional and nonfunctional tests on the system.The test results show that the system functions meet the expected requirements,the average response time of the interfaces is within 1.2s,and the system runs normally with the concurrency within 1w.At present,this system has been put into service,which has effectively improved the purchase experience of users and the experience of business personnel,improved the click rate and conversion rate of users,and enhanced the market competitiveness of the fresh food e-commerce platform.