Research On News Recommendation Algorithms Based On Text Processing

Due to the explosive growth of news information and the fast pace of life in modern society,users are increasingly demanding the accuracy of recommendations.In the field of personalized news recommendation,the ability to bring the news that the users are interested in by selecting from massive news quickly and accurately can enhance the users’ satisfactory,which have great business value.This thesis is based on text semantic mining to build a model for news text content and user interests,which builds news portrait and user portrait respectively.Firstly,based on empirical to compare and analyse three methods of text keyword extraction.The word frequency-inverse text frequency and space vector model are used to vectorise the text.Therefore,the keyword information in the text is fully extracted.Secondly,applied encoder to compress modelling the content,this allowed us to obtain low-dimensional vector representation of text while preserving original information,and solves the problem of text feature sparsity.The validity and accuracy of the model training are improved.Then,based on the user history browsing text information as input.Following,the time decay model,the Recurrent network model,the Long-short Term memory model and the Gated Recurrent model are respectively implemented to generate the interest vector of the target users;The Recurrent network model can spontaneously and iteratively update the user’s historical interest.The model can quickly capture the changes in the user’s reading interest and can express the user’s interest well.Finally,the similarity between the user and the article is calculated to select the similarity.The former k news are personalized recommendation for the target users.Based on the data of www.caixin.com,it can be seen that the length of the recommend-dation list has a great influence on the recommendation effect.Compared with the collaborative filtering method,the method adopted in the paper enables the recommendation accuracy to be enhanced by 2%and improves the recommendation diversity by 7%;compared with the content-based recommendation method,the method used in the thesis improves the recommendation accuracy by 13%and the recommendation diversity by 8%.At the same time,compared to the time decay model the method of recursive neural network modelling for users has an accuracy of 8%and a 5%increase in diversity.Additionally,the long-term and short-term memory models have the best performance,which verifies the effectiveness of the proposed method in this thesis.