Modeling and Characterizing Social Media Topics Using the Gamma Distribution

Abstract

We present a novel technique to identify emerging or important topics mentioned on social media. A sudden increase in related posts can indicate an occurrence of an external event. Assuming that the sequence of posts is a homogeneous Poisson process, this sudden change can be modeled using the Gamma distribution. Our Gamma curve fitter is used to return a set of emerging topics. We demonstrate our algorithm on Twitter data and evaluate empirically using the Reuters News Archive and manual inspection. Our experimental results show that our algorithm provides a good picture of the emerging topics discussed on Twitter.