Real-time recommendation for microblogs

Existing microblogging systems, such as Twitter, provide global or local discussion trends, so that users can easily find hot topics to follow on. The discussion trends are discovered by analyzing statistics of the microblogging database globally or regionally. As a consequence, users who select the same region, receive the same set of topics irrespective of their interests. This strategy fails to deliver relevant recommendations to the right users. To address this problem, in this paper, we propose a real-time customized recommendation scheme for microblogging systems. Our approach identifies users' interests via their personal tags and builds tag-user graphs for computing the similarities between microblogs and users. The subsequently submitted microblogs are considered as an input stream that flows through our tag-user graphs and buffered for the interested users. The user's browser can update his/her recommendations in real-time by pulling microblogs from our buffer. A statistics-based pruning approach, called APS (Approximate Pruning Scheme), is applied to reduce the processing cost by effectively avoiding unnecessary comparisons. We evaluate our system with two real datasets, namely the Twitter dataset and the Netease dataset. Our extensive experimental study shows the scalability and efficiency of our approach.