Research On Personalized News Recommendation Algorithm Based On Word Embedding

With the development of Internet technology,more and more people are turning their reading way into online reading.News is recommended as a means of positioning a product filter and the user,can according to user's history of reading habits for users to recommend the news might be interested topic,help users in the era of big data,accurate access to effective information,save a lot of reading cost,and effectively solve the information overload of the problems of large data.At present,the research on recommendation system under the condition of information overload mainly focuses on the field of e-commerce,and there are few researches on personalized recommendation of news.However,news,as an indispensable part of daily life,has a fast update and iteration speed and a large amount of information,making it difficult for users to timely capture the information they are interested in.Therefore,personalized recommendation for news is of great significance.The collaborative filtering model is a classic recommendation algorithm model,and has achieved good results.With the development of statistics,the neural network synergy filtering model that combines neural network and collaborative filtering has further improved the accuracy of the recommendation system.However,these two models are there are some limitations,such as by user interaction and project records to extract the user's behavior characteristic,did not make full use of the available additional information,recommended limits the accuracy of ascension,and sensitive for sparse data.Therefore,this paper will study the application of text mining technology and neural collaborative filtering model in the field of news recommendation.Firstly,the main recommendation algorithms and the related principles of text feature representation are analyzed and introduced in detail,and the advantages and disadvantages of each method are summarized.On this basis,combined with word embedding technology,neural collaborative filtering is improved.The specific work of this paper is as follows:(1)This paper firstly summarizes the current research status of recommendation algorithms and word vectors at home and abroad,and analyzes in detail the relevant principles,advantages and disadvantages of commonly used recommendation algorithms in the recommendation system,so as to provide theoretical reference for further research on recommendation algorithms.(2)To improve the neural collaborative filtering algorithm,aiming at the shortage of the model using only users interact with the news information,introduce the word embedded technology model,extract the characteristics of news,on the basis of user interaction and news information,news headlines,news content information as input characteristics of the model,to improve the model accuracy.(3)This paper makes a comparative analysis and research on the effects of three mainstream word embedding technologies,such as word-level word embedding model Word2 Vec,character-level word embedding model FastText and paragraph level word embedding model Doc2 Vec,on the personalized recommendation of Chinese news,and analyzes their advantages and disadvantages.Finally,this paper compares the recommendation effect of the optimized recommendation algorithm in this paper with that of traditional argot model and neural collaborative filtering model.The experimental results show that the optimized algorithm in this paper can improve the quality of recommendation system to a certain extent and effectively complete news recommendation tasks.