This data set contains aggregated user-level information about 491 Twitter users with their self-reported age and gender and dark triad scores obtained using the Dirty Dozen questionnaire.

Associated paper, read for more details:
Daniel Preotiuc-Pietro, Jordan Carpenter, Salvatore Giorgi, Lyle Ungar
Studying the Dark Triad of Personality using Twitter Behavior
CIKM 2016

For other questions about the dataset, contact Daniel Preotiuc-Pietro (www.preotiuc.ro)

Each feature file contain one feature category (space separated, header matches paper denomination):

feature-unigrams - tokens used by at least 1% of the users
feature-emotions - Ekman's six emotions + anticipation + trust + positive/negative computed using the NRC emotion lexicon
feature-LIWC2015 - LIWC 2015 categories
feature-w2v_200 - word2vec hard clusters, available from http://www.sas.upenn.edu/~danielpr/clusters.tar.gz
feature-image - profile image features
features-profile1 - profile features
features-profile2 - profile features (derived from tweets)
features-shallow - shallow tweet features
outcomes - user level information, 'ag' suffix indicates residual values after predicting using age and gender

Missing features are represented with -1
We extracted text features only for the 491 users that posted more than 500 tokens.
Shallow features and a part of the profile features were only extracted from the 710 users posted at least one tweet.

If using this data set, please cite the following publication:

@inproceedings {darktriad2016cikm,
	title = {{Studying the Dark Triad of Personality using Twitter Behavior}},
	author = {Preo\c{t}iuc-Pietro, Daniel and Carpenter, Jordan and Giorgi, Salvatore and Ungar, Lyle},
	series = {CIKM},
	booktitle = {Proceedings of the 25th {ACM} Conference on Information and Knowledge Management},
	year = {2016},
}