If the Twitter promotion algorithm works using machine learning, its code (not including the neural net) won't tell people much about how Twitter chooses what to show them. It also may not be much use to anyone.

Basically, the trained neural net is a part of the program. It can be considered source code, since (1) it isn't compiled from any other form of the code, and (2) changing it is possible (by retraining. But if you don't get that, you don't have the whole source code.