The Research And Implementation On Personalized News Recommendation Based On Storm

With the further development of the Internet, and mobile devices are expected to exceed 3 billion global Internet users in 2015. Internet has been to promote the era of big data, while a variety of information on the Internet is extremely expansion and redundancy, personalized recommended is particularly important in the era of big data. By calculating big data, recommender system can send users content that user will be interested in intelligently, so that people free from the confused mass data, and improve efficiencyCurrent news recommendation system based on collaborative filtering and recommended algorithm based content. But it faces the shortcomings of collaborative filtering, the sparsity of data and cold start defects.The main work of this paper is:1) Based collaborative filtering recommendationIn addition to the list which user access, the access list of news items as an external measure of similarity, we also joined the news keyword and news category as the calculation of similarity. Besides, according to the news published in time, we optimize parameters to the recommended values of recommended projects, with the recommended results are more in line with the user’s browsing behavior.2) Based content recommendationNews content includes keywords and news category.In the recommendation based on the news category, the number of news category, that user browses on, is positively related with how users interested in news of that category. After system analyzes each user’s browsing history, and count the number of each news category user browses, this number can be considered the value of user interest for each news category.In the recommendation based on the keywords, system calculates keyword vector for each user from history. Algorithm use the keyword vector as points that user is interested in. And system uses TF-IDF to recommend news set for users based on keyword vector. TF-IDF will calculates score for each item in recommendation set, and set news item in descending order according to score.Experimental results show: the final score of news is the sum of the score calculated by CF and the score calculated by algorithm based content, this way to mixed two algorithm can improve the recommendation set effectively.In order to spend up the computational process, system treats storm as core, and pushes content of each news and result calculated by Mapreduce to redis database, storm just recommend the result saved in redis.