Design And Implementation Of Personalized News Recommendation System Based On Improved SVD Algorithm

With the rapid development of informationization and popularity of Internet, the Web is continuously changing the way people access information. Now via the Web, people can get the latest news, read online books, and even purchase items, and so on. Almost all real-life needs can be met via the Internet.On the one hand, the rapid development of technology allows people to get information from the exceptionally rich sources of information. On the other hand, while people continuously digitized everything in real-life, we're still keeping producing massive UGC content, which formed a very considerable amount of digital information and will lead to the information overload problem. In other words, there is so much information that users can't adapt. It is difficult for ordinary people to efficiently find useful content on their own from such a mass of information.As an important channel for people to obtain information, News is also facing this problem. Whether it is traditional news portal, or navigation sites providing classified news articles, or recently active self-media, although all of those further enrich channels people get information from, there is no doubt that this also increase the information overload problem. Consequently, people find it difficult to dig out some suitable news for themselves.So, personalized news recommendation system comes to solve this situation, by providing a brand new way to get news. By analyzing historical browsing behavior of users and a variety of recommendation techniques, personalized news recommendation system can automatically predict the user's preferences, thus select specific news to meet user's interest from a mass of news sources.Based on this background, this paper analyzes some common recommendation algorithms, including contend-based algorithms and collaborative filtering algorithms, introduces leading distributed memory computing framework – Spark, analyzes its internal principles, such as computing model and data storage mechanism, finally we briefly introduce its eco-system BDAS. On the basis of analyzing parallel algorithm on the Spark platform, this paper implement the SVD matrix decomposition algorithm, which can be executed in parallel. Then we implement a prototype system for personalized news recommendation. Finally, we carry out a series of experiments, including parameter tuning, correctness validation, as well as benefit validation which is gotten from parallel computing.