Design And Implementation Of Commodity Recommendation System Based On Streaming Computing

With the rapid development of the Internet and information technology,people increasingly like to purchase goods from the Internet.The rise of e-commerce has indeed met people's demand for convenient shopping to a large extent.However,when the number of goods tend to be massive,consumers have a sense of being unable to start with a wide range of goods.This is when the amount of information reaches a certain scale.There is information overload.From the point of view of the e-commerce website,how to make the majority of consumers like and buy their own products,especially to show the long-tailed products in front of the consumers in need;from the perspective of consumers,how to increase the number of products Under the circumstances,it quickly finds the goods it needs without being disturbed by noise information.The recommendation system is an important tool to solve this contradiction.Through the mining and analysis of historical data of the user,a user characteristic model is constructed from multiple dimensions such as gender,age,interest,and hobbies,and when the user next generates a behavior on the website,the user can be recommended according to the characteristic model.Most of the recommender systems are faced with real-time and cold-start problems,because calculating the user,item similarity matrix,or matrix decomposition requires off-line calculation and is time consuming.Cold start is recommended because the newly registered user or newly added product has not produced any behavior.Due to the lack of historical data,the correct feature model cannot be calculated,so accurate recommendation results cannot be given.There are three options for solving the recommended cold start.First,the user's age,gender,and region are recommended based on the user's registration information.Second,the user is guided to express their preferences and use the collected user interest labels to recommend.The third is to recommend the most popular products for the most recent period of time to users.There is also a third-party platform that uses the user's behavior record on a third-party platform as a recommendation basis.This article mainly adopts hot product recommendation and flow calculation to solve the recommended cold start problem.The gray test proves that this solution can effectively solve the recommended cold start and real time problems.In this paper,through the in-depth study of the existing recommendation system and streaming processing technology,this paper proposes a real-time recommendation scheme for streaming computing.By establishing a multi-level data analysis and processing platform,the contradiction between the time-consuming calculation of historical data and the real-time response of new data is effectively solved.The solution is based on Apache's open source Storm and Hadoop frameworks.It combines the technical features of Redis and Elastic Search to implement off-line computing,near-line computing,and online computing from the architecture level.The analysis and modeling of historical data are mainly completed in the off-line computing platform.The calculation of real-time interaction data of the user is mainly concentrated on the near-line computing platform.The integration of the recommendation results with the business rules and reordering is realized through the online computing platform.Inter-data transmission and sharing are implemented through Kafka message queues and Redis caching.The real-time product recommendation system strictly complies with the principle of high cohesion,low coupling,and extensible software system architecture.In the structure of the paper,firstly,the relevant technologies of the recommendation system are reviewed,and the research status and industry development of the recommended technologies at home and abroad are analyzed and compared.Then,the actual requirements of the recommendation system and the real-time and cold start of the existing recommendation system are analyzed.On the basis of diversity and other issues,the recommendation system architecture with real-time quick response to user behavior is summarized.Finally,the real-time commodity recommendation system architecture and its implementation methods are described.