". . . none of the data used was associated with a specific person’s Facebook account . . ."
Maybe the results weren't matched to names but you were still dicking around with 'specific users' and that's the main problem.
Specific people were part of an experiment that, by the standards of those conducting the research, was not conducted with 'informed consent'. They can point to T&Cs all they like but a long stream with generic language about 'improving services' cannot possibly be considered adequate for a psychological experiment.