Research On Information Recommendation Methods Of Microblog Media

With the continuous development of the Internet and information technology, especially the emergence of variety social media in recent years, people can more easily to get a variety of information posted by their friends and others in social media. These techniques basically meet the user’s demand for real-time access to information, but it also bring troubling information overload at the same time. Therefore, as the most effective solution to this problem, recommendation system attracts more and more attention of researchers and engineers. Currently, researchers have proposed a variety of methods to solve the problem of merchandise, books and other field’s recommendation and achieved good results. But, unlike these area’s recommendation problem, as information recommendation lack the user’s external feedback, so we can’t directly use those traditional methods to solve the problem of information recommendation.For the absence of explicit feedback informat ion to solve the problem of information overload which we are faced with, this paper attempts to use the user’s and their friends’ historical data to get the information about the user’s interest, and then use the way of personalized news recommendation an d personalized ranking of their information stream to solve the information overload problem. The main work as follows:Firstly, we construct a web spider to get the experimental data. As for the data sparseness problem faced by recommendation system, taking into account of the weibo user’s feature, we designed two data extension methods to extend the user’s data, and experimentally proved that these two methods can effectively alleviate the problem of data sparse.To solve the problem of news recommendation, we firstly attempted to use the traditional content-based recommendation approach to solve the problem, as the recommendation results are not satisfactory for the problem, then we used the logistic regression and the support vector machines model to des ign two news recommendation methods and analyzed the existing problems.After analyzed the Logistic Regression and Support Vector Machine’s limitation, we have chosen the learning to rank method to solve the problem of news recommendation. We proposed two news recommendation methods which were based on Bayesian optimization criterion and Rank SVM. In addition, we also proposed two methods to solve the dynamic change of user interest and recommendation novelty and diversity.For the problem of information stream overload faced by users, the paper analyzed the user’s multi-purpose and the different attributes of the microblog platform and then proposed an information stream ranking method which had merged multiple features. We can use this method to rank the in formation stream for people in order to achieve the purpose of helping user to get useful information effectively.