That doesn't seem too surprising to me. After working on some EEG projects in the past (coincidentally within gatech) coming up with meaningful correlations that generalize across multiple subjects is very nontrivial and for a number of features you're not going to find anything of much use.

Tweets also do seem to be a poor feature to try to track, but generating labels for anything regarding EEG related problems tends to be quite fickle and very subjective.