Research On Personalized News Recommendation Method Based On Comment Relevance Analysis

With the increasing popularity of online news,the number of news is increasing explosively.Therefore,how to help users find the news they are interested in from the massive news efficiently and accurately has become the main research hotspot of scholars at home and abroad.The traditional collaborative filtering news recommendation algorithm is based on the user's browsing records,but the user's browsing records can not directly indicate the user's preference for news.However,the relevance between users' comments and news can intuitively reflect users' interest in the news.How to classify user comments efficiently and accurately according to the relevance of comments and news is a new and challenging problem.Moreover,because of the informal characteristics of comment text and the problems of grammatical errors and punctuation abuse,the traditional model can not capture the correlation between news and comment at a deeper level.Based on the above problems,this dissertation constructs a news-comment relevance classification model BERTRNC based on BERT(Bidirectional Encoder Representation from Transformers)feature extraction.On this basis,a personalized news recommendation algorithm based on user comments is proposed.The main contributions of this dissertation are as follows:(1)This dissertation proposes a news-comment relevance classification model(BERTRNC)based on BERT feature extraction.The model uses the BERT model trained in the extensive semantic database to extract the feature vectors of news and comments,and inputs the extracted feature vectors into the full connection layer network to complete the classification task.Then,aiming at the difference between the relevance of news and comment,this dissertation proposes a multi-classification loss function based on the relevance of news-comment.At the same time,aiming at the brevity,colloquialism,and non-literariness of news comments,this dissertation makes a fine-tuning operation on the BERT model,so that the model proposed in this dissertation can extract more accurate text features from colloquial comments more efficiently.Finally,a large number of experiments show that the BERTRNC model has better performance than other benchmark models.(2)This dissertation proposes a personalized news recommendation algorithm based on user comments(UserCF-HLC).Firstly,the algorithm proposes a calculation method of user similarity,which integrates the influence of news popularity and the text characteristics of user comments.Then,this dissertation integrates the relevance between comments and news into the calculation of users' news preference,and measures the confidence of users' news browsing records according to the relevance level.Secondly,the core idea of collaborative filtering is applied to generate a news recommendation list that is more in line with users' interests.Finally,a large number of experiments are carried out.The experimental results show that the personalized news recommendation algorithm integrating user comments has significantly improved the recommendation effect compared with other algorithms.