As a historian, I work a lot with time-series data. And I’m always looking for methods from time series analysis to model my data in new ways. A while ago I came across Kleinberg’s burst algorithm, which is a great method for finding ‘trending topics’ in historical textual data. Although it can also be applied to other signals. This blog post by Nikki Marinsek provides a nice introduction to the algorithm.
Looks cool! Do you have any recommendations on setting the parameter s? Or is that a matter of trial and error?
It depends very much on the data, and the resolution of the time steps you are using, so in my experience it’s mostly trial and error.