I guess it sort of depends on what you mean by the algorithm. I'm not an expert on this either, but the general idea is that you pick some sort of thing you want to promote (e.g. engagement), and then you use a really large data set with thousands of parameters to construct the `parameters -> engagement function`, some of those parameters will be user supplied, and some will be Facebook settings.

The thing that's opaque is this function that the program produces. You can really only know how the function works experimentally.