A Personalized News Recommendation Research Based On Lda2vec And Restricted Boltzmann Machine

As news websites or APPs have wider user coverage,so if users can access their favorite news as easily as possible,and reduce the cost of getting news,will result in a greater social and commercial value.This is the meaning of personalized recommendation techniques apply to the news system.News field,however,there are several obvious characteristics,one is the timeliness of news and high speed,the second is very serious cold start problems in news recommendation,three is the users in the field-readers,more easily influenced by popular and hot news.Due to the above factors,make personalized news recommendation system is different from the traditional personalized recommendation.In the past,method based on content is commonly used in news recommendation to get the long-term interest of users,however,which used to the extract content information often cannot capture precise semantic information,which leads to poor recommendations.When use the collaborative filtering,because of the problem of cold start,it is difficult to recommend news to users in time.This paper studies how to implement the personalized news recommendation system with the effect and performance.We proposes a personalized news recommendation model based on lda2vec and Restricted Boltzmann Machine(RBM)in this paper.The model can extract the theme information of news,and according to the conversion rules,press the user implicit behavior data into user's explicit rating data,then news topic information and the user's rating data respectively used as a condition layer and visual layer of this model,finally realize news recommendation based on conditional RBM.In view of the news text,this paper proposes a clustering method based on lda2vec to realize the topic extraction.At the same time,in order to validate the accuracy of the recommendation framework,we conducted a series of comparative experiments,such as clustering algorithm based on lda2vec and word2vec?LDA algorithm,and comparing the news recommendation framework proposed in this paper with other collaborative filtering models,such as item-based CF,SVD,and comparing the news recommendation framework proposed in this paper with the framework based on word2vec and RBM?the framework based on LDA and RBM.Experimental results show that the news recommendation model proposed in this paper is feasible.