The "algorithm" isn't some simple KNN. It has many people working on it, so by now it should at least grasp the difference between two topics that are similar and two topics that are the same.