Research And Application Of Intelligent User Recommender Systems In Microblog

In recent years, with the rapid development of social networking sites, almost every internet user is using different social networking sites and services based on them every day. Social relationship is the most important data of social networking sites and the foundation of all other services, so the user recommender systems in social networking sites has a very important research significance and application value. Base on the data from Chinese famous social networking site named Tencent microblog, we have an in-depth study of the collaborative filtering algorithms. The specific studies are as follows.(1) We propose the main problems of microblog’s user recommendation systems based on data analysis; implement three baseline models which provide experimental basis for the later improvement of collaborative filtering algorithms.(2) We improve the traditional user-based and item-based neighbor models by utilizing the social relations, directories and the keywords.(3) We propose an asymmetric factor model to integrate the social relations which solve the cold start problem and make a great improvement in the prediction accuracy; then based on this model we integrates the context information to further improve the accuracy.(4) We implement two combined models named linear regression combination and neural network combination, by combining multiple models’prediction scores; the experiments show that integration of the various types of information in a single model is a better way to improve the prediction accuracy than combination of multiple models to use these information.