A Refined Method for Detecting Interpretable and Real-Time Bursty Topic in Microblog Stream

The real-time detection of bursty topics on microblog has acquired much research efforts in recent years, due to its wide use in a range of user-focused tasks such as information recommendation, trend analysis, and document search. Most existing methods can achieve good performance on real-time detection, but unfortunately, lack of much consideration on topic coherence and topic granularity for better semantic interpretability, which often results in odd topics hard to be interpreted. Therefore, it demands much more efforts on evaluation and improvement of the intrinsic quality of detected topics at their very early stages. In this paper, we propose a refined tensor decomposition model to effectively detect bursty topics, and at the same time, evaluate topic coherence and provide informative bursty topics with different burst levels. We evaluated our method over 7 million microblog stream. The experiment results demonstrate both efficiency in topic detection and effectiveness in topic interpretability. Specifically, our method on a single machine can consistently handle millions of microblogs per day and present ranked interpretable topics with different burst levels.