Design And Implementation Of Microblogging Recommendation System Based On Collaborative Filtering

The microblogging system allows users to publish brief messages shorter than140words，and has been added SNS feature. In the last three years, there are more than200million demotic registered users, and the microblogging system has become the mostpopular information publishing and sharing platform. So many users generate vast amountof data, and this leads to the information overload phenomenon. There is an urgent needfor a personalized service system which can learn users’ interest, filter and recommendinformation based on users’ interest.This thesis focuses on how to recommend microblogging information. According tothe requirements, we choose the collaborative filtering method, design and implement amicroblogging recommendation system based on this method. The system mainly includesthree parts:1) basic data collection,2) recommendation result calculation, and3) userinteraction. In the first part, data is collected from sina microblogging system by open API,and taken as the source of recommendation. In the second part, user’s interests areclassified, and the interest model is built though the explicit and implicit user feedback.Then in accordance with the users’ interest model, the collaborativefiltering recommendation engine filters the information which may interest users from thebasic data, and creates the recommended results. In the last part, users can login in the website and get the recommended information. At the same time, they can feedback to thesystem, which can help the system adjust their interest model, making therecommendation results better. According to the test result, we can conclude that thesystem can recommend accurate results, and has better stability.The system has passed the examination of Sina Company and has been deployed onthe real online environment. Through analysis of the users’ actual experience, the systemcan generate good recommendation results in most interest areas, and can be fundament tosolve the information overload problem.