Abstract:

Due to fast update of news,the clustering- based preprocessing is usually needed when the news is recom-mended to users.However,some traditional clustering methods are too complicated while others rely on iterative ini-tial value,none of which can be accurately and effectively applied to news recommendation.Considering the aboveissues,we propose a news recommendation method based on two- fold clustering.In this method,a density clusteringof random sample data is conducted.Based on the cluster number and initial cluster center of the density clustering,a fast two- fold clustering of all the news to be recommended is performed.Then,the news recommendation is realizedby combining such factors as fashionability and popularity.The proposed method can cluster relevant news without toomuch computation cost,and it can calculate parameters by means of parameter estimation.Experimental results showthat the proposed method is superior to other news recommendation methods in terms of diversity and accuracy.

Key words: news recommendation, text clustering, text processing, personalized recommendation