The Design And Implementation Of Takeaway Product Recommendation Model Management And Monitoring System For Mobile Terminals

With the rapid development of mobile Internet technology,the amount of informa-tion on network platforms has increased exponentially.Especially since the outbreak of the epidemic,the offline consumption scenes based on store have been seriously damaged,and people's consumption concepts have gradually changed to more conve-nient online consumption.In order to capture user consumption trends and improve corporate economic benefits,most Internet companies often implant interactive rec-ommendation elements in user terminals.However,for recommendation models,the core of recommendation decision-making,the industry lacks a complete and efficient management and monitoring tool to cope with scenarios where models are launched on multiple terminals or to avoid risks caused by models that fail to meet expectations after launched.In response to the above problems,this thesis designs and implements a takeaway product recommendation model management and monitoring system for mobile ter-minals,which mainly serves the company's takeaway business and aims to help algo-rithm developers effectively manage and monitor the whole process of model training,publish,and recall,and make timely intervention when the online effect fails to meet expectations.The system collects user gesture data and commodity data,filters invalid gesture data with the help of ”Douglas-Peucker” algorithm,and distributes the charac-teristic data to algorithm developers through Redis.This system converts the original models constructed and trained by algorithm developers into models that can run on Android and iOS systems and publishes them to user terminals.This system provides a complete visual monitoring mechanism to help algorithm developers perceive the op-erating status of online models in real time.In addition,if the effect of a certain version of the model is not as expected after it is launched,the system will activate the early warning or recall mechanism,automatically alert or recall the model and notify the algorithm developers.This system is based on the React framework and Spring Boot framework to im-plement front-end Web platform and back-end business services,and maintains good scalability of the system.The front-end uses ECharts components to realize data visu-alization.The back-end can be divided into four modules: feature management,model management,publish management,and model monitoring.Each module uses mea-sures such as current limiting fuse to ensure system availability,and provides external services through RPC.At present,the system has been deployed to the company's private cloud platform and is running online,and algorithm developers can log in to the system through their personal MIS accounts.Long-term trial operation shows that this system has signifi-cantly improved the efficiency of model development and publishment for algorithm developers,and automatically triggers the early warning or recall mechanism when a certain version of the model does not meet expectations,so as to effectively avoid the risk of online production and meet the delivery expectations.