Table 3. Cutoff threshold and corresponding number of unique values of candidate features CF for learning. Thresholds were chosen to balance the number of each type of feature.

	Frequency threshold	#Unique values
User	235	206,084
Hashtag	65	201,204
Mention	230	200,051
Location	160	205,884
Term	200	204,712
Total candidate
Features (CF)	–	1,017,935