Study And Implementation Of News Recommendation System Based On Topic Model

Today, a large part of users receives the latest information through the internet, although the network information is in a state of overload. Since the traditional news sites just push the popular information to users, this will lead to an unhappy result, i.e. everyone gets the same information, which means that users cannot get their own real-concerned information from the news. To solve the problem, the information recommendation system using recommendation algorithm recommends news by mining user's preferences in the historical data of users' behavior. Therefore, personalized information recommendation method also received a lot of attention.This paper calculates the distribution of the user's interest topic model based on user topic model, and then completes the model of the news text content by using the topic modeling method. After calculating the distribution of the article topic, content-based recommendations algorithm is used to recommend the articles to users. The recommended articles are the similar articles between user topic distribution and article topic distribution. The research works are as follows:(1) Firstly, the key algorithms in personalized recommendation and the related technologies were analyzed and compared; then the content-based recommendations algorithm used in this system was described.(2) The system described in this paper is the personalized recommendation system based on mobile user profile. The paper analyzed the interest characteristics of each user in news information according to the user profile and established the user interest model; then the topic distribution of users was calculated in this paper. On the other hand, via adopting the LDA(Latent Dirichlet Allocation) topic model, this paper used a large number of news information to get the topic of articles by machine learning and established the topic model. Finally this paper calculated the topic distribution of articles which are similar between the user's topic distribution and articles' distribution. Then these articles were recommended. The recommended methods achieved by using weighted sorting algorithm to update users' recommend list. Offline and online modules were designed to predict inactive and active interests in the distribution of users, so that the modules could make recommendations for these users timely.(3) A personalized information recommendation system was designed and implemented in this paper. The performance and functionality were tested to verify the function and performance of the system. The results showed that the system illustrated in this paper promoted the performance of the recommendation system.