Research Of Personalized News Recommendation Algorithm Based On Topic Models And User's Interest

Online news reading has become a popular way to read news articles from a huge collection of news sources around the world. News recommendation systems help users manage this flood by suggesting articles based on user interests rather than showing news articles in the order of their occurrence. News reading services, like Google News and Yahoo! News, have become increasingly prevalent and known by people as the Internet provides fast access to news articles from different kind of information sources around the world. As news articles every day have large number of production, one of the biggest problem that online news provider need to face is how to make full use of the user information and news articles information to help users find as much as possible news articles which are similar to user reading preference, recommending news articles has become a promising research direction.Although personalized news recommendation have made some new achievements, but still remains challenging for at least the following several reasons. First, the popularity and the recency of news articles change dramatically over time, which differentiates news items from other objects, such as products and movies; Second, news articles which read are not independent in the majority situation, in other words, reading one news articles may affect the subsequent news articles reading; Third, the latent relationships among different news articles, and the special properties of news articles didn't get fully utilized. In a word,many critical issues of news recommendation have not been efficiently resolved in previous studies.In order to solve the above problem, this paper analysis and research to the traditional recommendation algorithm, a personalized news recommendation system framework is presented. The main research include the following several aspects:(1) In order to solve the cold start problem and data sparsity problem of the present recommendation algorithm and insufficient of the similarity computation algorithm, we fully analyzes the latent special properties of news articles, and to explore the possibility of integration of different latent special properties of news articles, an improved similarity computation algorithm are proposed based on integration of news articles, user reading behavior and the named entity of news articles. The effect of the time hot effect and less popular news articles on the result of recommendation also have been considered in this algorithm.(2) We intensive analysis the effect of user interest evolution on the result of recommendation when recommending news articles, and represent the user reading preference with integration of long-term and short-term user interest.(3) We choose a two-stage news selection strategy based on long-term and short-term user interest. The long-term user interest is firstly utilized to differentiate news groups according to the clustering algorithm divided with specified preference, and then the short-term user interest is applied to filter and recommend specific news articles to individual users from each news group.(4) We research how to realize the fusion of the two methods based on clustering algorithm and the user interest model to produce a novel framework for news recommendation.Finally simulation experiment on the data set and analyzing experimental results, the experiments show that the news recommendation algorithm based on latent properties of news articles and user interest evolution compared with the traditional news recommendation algorithm can improve the accuracy of news recommendation system.