It is bunch of heuristics... just slightly more of them producing (in my opinion) slightly better result.

The core of the article however points at the main problem: no evaluation dataset on which we could compare these algorithms.