Datasets for link prediction? can you clarify your question a bit more?
if what your looking is random generation of links metrics, then it a matter of link quality and it depends on you how do want to quantify the link quality. and it can be applied in graph theory in general.
I hope this helps. if you can elaborate more about your issue it might give better answers from other experts.
take for instance, I have a time shot of facebook graph at time t1, and a next shot at time t2. Precisely I want these shots with multiple time stamps.