A Real-time News Recommendation System Based On Subject Extraction & Evolution Model

With the rapid development of Internet, more and more people use internet to get news information, in the face of huge amounts of news and information,information overload problem will makes people confused, personalized news recommendation system can solve the problem of news and information overload. News recommendation system can realize real-time automatic processing of journalism, and through the news topic evolution extraction algorithm, combined with the user, relate news content and users, using real-time data stream processing technology, can solve the problem of big data under the background of the news accuracy recommended.In this paper, we designed an evolution of news subject extraction method based on real-time recommendation system, main work includes:(1) Based on classical LDA (Latent Dirichlet Allocation) algorithm,we designed news subject extraction and topic evolution of LDA algorithm. LDA algorithm combining with the algorithm, increased the time evolution of the deductive process, by building a cascade news and time table to calculate the independence test chi-square parameters, can keep track of the time evolution of the news events more features, and more effective to extract the theme of the news; Using distributed indexing technology based on solrcloud news text, high performance distributed computing and experimental simulation show that system has good performance.(2)We designed based on a user portrait of alternating least squares UP-ALS collaborative recommendation algorithm. New algorithm combining the user portrait label and alternating least squares ALS prediction score matrix to complete the calculation of collaborative recommendation, more accurate positioning user preferences; To meet the need of real-time computing, big data environment Spark distributed stream processing framework is adopted to accomplish the UP-ALS algorithm; Experimental simulation show that compared with the model MFM and ALS matrix decomposition algorithm, the UP-ALS algorithm improved the accuracy of recommendation.(3) The use of distributed architecture to build the real-time recommendation system, and to tune the performance of the system, set up the system in the cluster, show the final effect of system implementation, can meet the demand of the user’s personalized reading well.