This dataset features all the tweetids and labels that were used to model the language of 24 hashtags, and test the performance on predicting the hashtags in unseen tweets. This study is described in:

Kunneman, F.A., Liebrecht, C.C. & Bosch, A.P.J. van den (2014). The (Un)Predictability of Emotional Hashtags in Twitter. In Proceedings of the 5th Workshop on Language Analysis for Social Media (LASM) @ EACL 2014 (pp. 26-34). s.l.: Association for Computational Linguistics, http://hdl.handle.net/2066/127067

In addition to the train and test data, this dataset includes the most indicative features (words and phrases) for four of the hashtags, as well as the human judgement whether the tweets that contain or are classified with these hashtags convey the presumed emotion of the hashtags.

Subject period: December 16th 2010 until February 1st 2013