基于LFM算法的SpringCloud分布式购物系统

SpringCloud Distributed Shopping SystemBased on LFM Algorithm

数据个性化推荐缺乏、无意义流量对系统的影响,以及后期代码维护困难是目前在线购物平台面临的主要问题。基于隐语义模型(Latent Factor Model, LFM)和SpringCloud框架,实现了分布式购物系统。针对数据个性化推荐问题,通过数据实时同步工具Maxwell采集用户数据并推送至消息中间件RabbitMQ,日志收集系统Flume接收数据并存储至分布式架构Hadoop中,由推荐服务读取数据,基于LFM的协同过滤算法实现数据个性化推荐;针对系统冗余压力问题,利用微服务网关Zuul限制子系统路由访问量,以减少子系统的流量压力;针对代码维护问题,使用Zuul和Nacos注册中心实现动态路由代理,自动分发流量至新增服务器已注册的子系统中。上述技术提高了在线购物平台的拓展性和应对高并发流量的处理能力,降低了代码维护难度。

The main problems facing online shopping platforms are the lack of personal data recommendation,the impact of meaningless traffic on the system,and the difficulty of code maintenance in the later period.A distributed shopping system is implemented based on Latent Factor Model(LFM)and SpringCloud framework.The user data is collected through the real-time data synchronization tool Maxwell and pushed to the message-oriented middleware RabbitMQ.The log collection system Flume receives data and stores it in the distributed Hadoop architecture.The data is taken by the recommendation service.Finally,the LFM-based collaborative filtering algorithm realizes personalized data recommendation.In order to reduce the subsystem's traffic pressure,the micro-service gateway Zuul is used to limit the subsystem's routing access.Zuul and Nacos registries are used to implement dynamic routing proxies to automatically distribute traffic to the subsystem registered by the new server.In this way,the code is easy to maintain.Based on the above techniques,the extensibility of the online shopping platform is improved,the processing ability to deal with high concurrent traffic is guaranteed,and the difficulty of code maintenance is reduced.