Design And Implementation Of Enterprise Personalized News Recommendation System

In the era of information overload on the Internet,users want to see news that they care about conveniently in the face of a large number of news information every day.The news sytem platform hopes to push the content that users are interested in to the corresponding users in the daily mass news.It is of great economic and social significance to solve the problem of information overload.At present,personalized news recommendation systems are usually based on collaborative filtering or content-based recommendation methods.The traditional collaborative filtering recommendation methods usually have the problem of sparse matrix.However,content-based recommendation methods may have a single result,and the cold-start problem has a great influence.In application,it needs to choose suitable methods for landscape and effect or integrate them with certain methods.In this paper,a content recommendation and collaborative filtering fusion recommendation model based on interest label is proposed according to the needs and practices of the enterprise.This paper mainly completed the following work:(1)Build user interest label model base on label vector.The text is processed by extracting keywords,introducing the concept of word similarity,transforming the uniform label vector,building interest model according to reading behavior,and getting user interest label vector.Based on the above process,content recommendation is carried out to optimize the method of matching user interest directly by text keywords.(2)Clustering users based on user interest tag model,and then implementing collaborative filtering recommendations for user groups,which makes up for the problem that the results of content recommendation based on individual historical behavior may be the lack of variety.It also solves the problem of excessive computational complexity of direct collaborative filtering and the influence of sparse matrix on the results to a certain extent.(3)A fusion recommendation model is established,which combines label-based content recommendation with collaborative filtering recommendation results,and combines news warmth factors to solve the problem of cold-start users.Through the comprehensive model,the recommended effect is kept relatively stable under the background of requirement.(4)An enterprise level news recommendation system is designed and implemented.Hadoop platform distributed computing technology is used in text content processing,user clustering and other related modules to achieve efficient and scalable large-scale data processing.