Microblog bursty feature detection based on dynamics model

Microblog is becoming more and more popular in our life. Due to the numerous information on this platform, it is very useful to detect bursty topic in real-time to help people get essential information quickly. As a necessary stage, detecting busty feature effectively is important for bursty topic detection. Based on dynamics model, we propose a new microblog bursty feature detection method. Firstly, we compute term weight taking account of both term frequency and tweet weight, where tweet weight factors include retweet number, comments number and time fading factor. After computing all terms´ weight, a bursty feature detection method is proposed based on dynamics model. On the analogy of physical dynamics model, we compute each term´s momentum by using MACD (Moving Average Convergence/Divergence) and determine whether it is a bursty feature in a given time interval. We employ our method to detect the bursty terms of Sina tweets with a series of experiments. It is demonstrated that our method is able to detect bursts for news terms accurately and efficiently.