Microblogging Recommended System Collaborative Filtering And Behavioral Analysis

Recommender system is a useful tool for helping users discover content of their interests and overcome the problem of information overload. Social network is an important platform for the publishment, spread and acquisition of information. As both are hot research topic in academia and industry, combination of the two is a major trend.In the scenario of followee recommendation in social network, this paper does an in-depth research into the core features of recommender system and social network. Main work of this paper is listed as follows:1A survey of recommender system methodologies:This paper categorizes and sum-marizes existing theories and methods in recommender system including content fil-tering, collaborative filtering and social filtering, with an emphasis in state-of-the-art algorithms in rating prediction problem.2Recommendation algorithms combining learning to rank:Rating prediction is a fun-damental problem in recommender system, but a more common form in real-world applications is TopN recommendation. Based on several modeling methods in rating prediction, this paper tries to apply learning to rank to those models. Experimental results show that it outperforms traditional rating-based models significantly.3Modeling of user interests in social network:There is plenty of heterogeneous infor-mation in social network, such as content and social relationships, etc. This paper tries to incorporate all information available into user interests modeling and analyze their different contributions through experiments.4Analysis and modeling of user behavioral pattern in social network:The main idea of recommender system is to derive user interests from their historical behaviors, thus most studies are centered in user interests modeling while modeling with regard to user behavioral pattern itself is relatively scarce.3different ways of modeling user behavioral patterns are proposed in this paper and then combined with user interests models. Experimental results show that the ensemble models can improve recom-mendation performance effectively.