Middle-Aged And Elderly Personalized News Recommendation System Based On BERT Model

In recent years,the rapid development of the Internet has brought massive amounts of news and information.At the same time,with the gradual aggravation of aging in China,the middle-aged and elderly groups cannot quickly and effectively obtain the content they are interested in when faced with massive amounts of news and information.In addition,many news and information platforms have a large number of harassing advertisements,low-quality news and inductive content,which not only affect the reading experience of middle-aged and elderly people,but also increase their chances of being cheated easily.In order to solve the above problems,this thesis studies a personalized news recommendation system for the middle-aged and elderly.The system consists of four modules.The news content filtering module uses a Bidirectional Encoder Representations from Transformers(BERT)pre-training model to filter candidate news for recommendation,and the news subject classification module classifies news to obtain news categories,and the news recommendation module applies recommendation technology to recommend personalized news to users,and the news keyword extraction module is used to obtain the news keywords browsed by the user history.The research contents of this thesis are as follows:1.Filtered the news content and select the candidate news recommended to the middle-aged and elderly groups.First,the news data is deduplicated.Then filter sensitive words.Finally,the BERT classification model combined with multi-layer Perceptron(MLP)and Bi GRU proposed in this paper was used to filter junk news.The effectiveness of the algorithm is verified through experiments,and the effect of spam news classification is effectively improved.2.Build the news topic classification model and determine the categories of news.The news is classified by data processing,multi-label classification model is built by using BERT classification algorithm combining MLP and Bi GRU,news category prediction and other steps.Through experimental analysis,this thesis verifies the effectiveness of the algorithm in news classification and effectively improves the effect of news topic classification.3.Implement news recommendation model and push personalized news to users.A series of operations,such as data preprocessing,feature engineering,feature selection and x Deep FM model construction,are carried out to construct personalized recommendation model.At the same time,a feature selection method combining deep learning and machine learning is proposed in this thesis.Through comparative experiments,it is proved that this method can effectively improve the effect of feature selection.Finally,in the part of model construction,we compare different recommendation models and verify that XDeep FM model has better effect in personalized recommendation.4.Key words are extracted from the news content that users have browsed historically to grasp the real-time preferences of users.