Research On News Recommendation Based On Deep Learning

News recommendation is of great significance to alleviate the overload of news information,which makes it difficult for users to get the content they are interested in by browsing all the news.News recommendation provides the news that users may be interested in through active recommendation service,which is especially suitable for the situation that users have no clear intention or cannot accurately describe their own subconscious needs.In recent years,the continuous innovation of deep learning technology promotes the rapid development of the field of news recommendation,which has been widely concerned by many researchers.Firstly,research status of news recommendation based on deep learning is classified and reviewed in this paper.The modeling procedures,the model structure,advantages and limitations of the representative algorithm models are expounded.Secondly,inspired by the hot factors in traditional news recommendation,hot factors are introduced into the deep learning news recommendation framework in this paper,and a hotspot-aware attention enhanced news recommendation model is proposed,which includes four components: news encoder,hotspot feature extractor,user interest extractor and click predictor.In order to improve the expression ability of text features,a news encoder with self-attention network is designed to extract discontinuous news features;in order to effectively mine hotspot features,a hotspot feature extractor is designed,which uses attention network to dynamically aggregate hot news and learn hotspot feature representation;finally,in order to improve the accuracy of predicting the click probability of candidate news,a novel click predictor is proposed to flexibly fuse hotspot features,user interest feature and candidate news representation.A real-world news recommendation dataset is also constructed in this paper,which is collected from the Green News website of Beijing Forestry University.A large number of experiments on the dataset show that our model has good performance in news recommendation,and the AUC and F1 increase by 7.51% and 8.63% respectively.At the same time,the model also helps to alleviate the cold-start problem of users.Finally,a news recommendation prototype system for Green News website is constructed based on our model.The system can provide users with a list of news recommendations that they are interested in based on the website hotspots and users' click history of Green News website,and it shows the application effect of the model visually.