Research On News Recommendation Methods Based On Event Topic Analysis

With the developing of internet,abundant of news information has been produced.However,since the news data is so large,the problem of 'information overload' has appeared.users often waste plenty of time in searching news,they are so difficult to find out what they are interested in.So that research on automatic recommending interested news to users is very necessary,and used data mining and machine learning technology to find out what the users like is very important.The news recommendation methods existed were rarely considered the event information to the recommendation processing.In this paper,we will do some research about news topic constructed,and news recommended method combining users' ratings and news topic model.the main achievements are as follows:(1)Firstly,news page on the internet were collected by using web spiders.Patterns were defined for parsing the title,release time,source,body of news.News database was constructed.We randomly assigned news document to the users,and got the ratings that users to the news that have read,and the users to news rating matrix used as a basic work of news recommendation.(2)Secondly,we proposed a news topic model construct method based on event element.Getting the correlative relationship among event sentences by analyzing the element in the event sentence,and make the correlative relationship as the direction constraint in the generating processing of LDA topic model.Thus more accurate probability distribution in the topic space of news document was obtained,as a foundation of news recommendation.(3)Thirdly,An collaborate filter news recommendation algorithm combing with topic model is proposed.The user-news rating matrix and news topic distribution are combined into the processing of matrix decomposition,at last get the rating of user to news that user haven't read,and recommend the news with high rating to the user.