Research On Personalized Service Of Recommendation System Based On Distributed Platform

The development of the Internet leaves a profound impact on enterprises and personal life. With wide use and popularization of Web, the application of ecommerce transactions had become increasingly popular. One of the most typical problems in the application of network electronic commerce is the "information overload", in order to help customers to better choose the goods on the Internet when confronted with a wide variety of Internet information and a large number of goods, the recommendation system emerged as required. Recommendation is a strategy to mining the information of user interests in large scale data, but with the continuous rapid growth of the scope of electronic commerce system, the implementation of the recommendation system is also faced with a series of challenges, which affect the development mostly comes from the following issues: feature extraction conundrum, cold start conundrum, data sparsity conundrum, extension of algorithm conundrum, etc.However, the recommendation system deployed with single computer can not deal with the massive scale of data, so it is a need to transform and improve the traditional recommendation system to adjust to distributed platform based on using cloud platform to deal with massive scale data, which is in the trend of the future development of the recommendation system.Based on the above background and the latest researches, this study focus on the core principles and deficiencies of several most popular used recommended algorithms in the industry application, such as: collaborative filtering algorithm, slopeone algorithm, content based recommendation algorithm, etc. Trying to use the combined recommendation model to fuse the advantages of different recommendation technology to maximize the recommended model from the point of view of user's and the item's characteristics respectively. In order to be able to deal with recommedation applications in large scale, In this paper, we attempt to combine the improved recommendation methods with the high efficiency of parallel computing power of Hadoop platform to achieve a distributed recommendation system and make a comparison between the proposed algorithms and the results of latest theory research after carrying out a lot of simulation experiments. The result shows that the promoted methods are more efficient and accurate than the traditional recommendation algorithm, and has good scalability, and can be used in the simulated recommendation service application.Finally in this study, we design a distributed personalized recommendation system composed of three parts: Web application layer, recommendation engine and database storage layer, based on the distributed recommendation algorithm. The system can ensure high precision and fast response to user needs by separating the online recommendation and offline calculation process.