Research On News Recommendation Method Based On Collaborative Filtering

In an era of information,a large amount of news will be generated every day,which cause the problem of information overload.It is difficult for users to obtain the content they really care about from a lot of news.The recommendation system can solve the problem efficient.As a classic method in the field of recommendation systems,collaborative filtering has been widely studied and applied.In this dissertation,a news recommendation algorithm based on collaborative filtering is established,which comprehensively considers the similarity with the user’s news reading history and news titles.And then,proposing a more effective user similarity compute method,which can improve the effect of news recommendation.First,in this dissertation,we propose a news title similarity computation method.Because the features of news titles are sparse,traditional similarity computation methods can’t effectively solve this problem.We propose a news title similarity computation method based on feature expansion and Siamese network.Firstly,a LDA model is constructed to expand the features of a short text.Then,deep features are extracted by using Siamese neural networks model which contains both CNN and Bi LSTM.Finally,the similarity of two news titles can be achieved by computing the Manhattan distance between generated feature vectors of these two titles.Experimental results show that,our method achieves higher accuracy and F1 score.After that,this dissertation proposes a news recommendation algorithm that integrates user reading records and the similarity of news title.This method comprehensively considers the user’s news reading record,the corresponding reading time and the similarity of the news titles to compute the similarity between different users.After that,the collaborative filtering is applied to recommend news that the target user may be interest in.Through experimental,the algorithm in this dissertation has achieved better results in precision,recall and F1 score.