Design And Implementation Of Data Distribution Subsystem Of Real-time Recommendation System

With the development of Internet technology,the amount of information grows,users are unable to find useful information on their own in a lot of information,and it is therefore recommendation system came into being.Recommendation system users and items will connect via different media,the media can be items that the user liked,or some other users who have similar interests with the user,or semantic with which the user describe items.Offline recommendation is the main body of traditional personalized recommendation system.It analyzes the data,updates the model,and then uses the new model to personalize recommendation.Because it is regularly updated model,the model cannot maintain real-time,the results of recommendation may not be very accurate.The great advantage of real-time recommendation is the enhanced accuracy of recommendation for users,and also that the users can quickly feedback based on the results of real-time recommendation,and then the recommendation model can be updated as soon as possible,so that it can be more accurate.The key question of personalized recommendation system architecture is how to achieve a seamless way to implement the calculation process of online and offline,to reasonably allocate the steps of the algorithm to the various parts,so that the resulting model can not only be very accurate,but also be very efficient to meet the real-time requirements.This thesis studies the issues related to data communication between the web portal and the recommended computing systems and proposes the data distribution subsystem of recommendation system.The subsystem meets high concurrency and real-time requirements from the web portal.The solution can provide recommendations of the data source for different service needs for the web portal and supports both offline and online recommendation service.This thesis design and implement the data distribution subsystem of recommendation system based on the Chinese mobile phone recommended reading service.Open source framework and interface technology are combined to achieve MVC framework model.Struts and Spring-based lightweight frame makes the system have good scalability,maintainability.Based on service requirements analysis,system design,the system is divided into six modules and realized:load balance,authentication management,offline recommendation service processes,online recommendation service processes,recommend books cache and user behavior information process.Finally,the functionality and non-functionality of the system are tested.The system ultimately implements all functional requirements and the performance requirement that supports 500 million concurrent requests per day and the average response time of less than 300ms.