1. Title: DEtection of TOXicity in comments In Spanish (DETOXIS)

2. Organizers: 

	Mariona Taulé 		mtaule[at]ub.edu
	Montserrat Nofre 	montsenofre[at]ub.edu
	Alejandro Ariza 	alejandro.ariza14[at]ub.edu
	Enrique Amigó 		enrique[at]lsi.uned.es
	Paolo Rosso 		prosso[at]dsic.upv.es

3. Corpus Description: https://detoxisiberlef.wixsite.com/website/corpus

4. Number of training instances: 3463

5. Number of attributes (including target labels): 21

6. Attribute information:

a) topic 			(categorical attribute with three possible values: CR = crime, MI = migration, SO = social)
b) thread_id 		(identifier of each conversation thread)
c) comment_id
d) reply_to 		(identifier of the previous comment in the thread. It defaults to itself in case it is the first comment in the thread)
e) comment_level	(categorical attribute with two possible values: 1 = direct comment to an article, 2 = reply to a comment)
f) comment 			
g) argumentation	
h) constructiveness	
i) positive_stance	
j) negative_stance	
k) target_person	
l) target_group		
m) stereotype		
n) sarcasm			
o) mockery			
p) insult			
q) improper_language
r) aggressiveness	
s) intolerance		
t) toxicity 		(TARGET attribute with two possible values: 0 = not toxic, 1 = toxic)
u) toxicity_level	(TARGET attribute with four possible values: 0 = not toxic, 1 = mildly toxic, 2 = toxic, 3 = very toxic)

Attributes g)-s) are binary features with value 0 if it is not satisfied given the respective comment or 1 otherwise. Furthermore, this set of features together with the topic attribute will only be available during training. The test dataset will only contain the following attributes: [thread_id, comment_id, reply_to, comment_level, comment].

There are no missing values in this dataset.

7. File format: both train and test files will be provided in csv format where each field is comma separated.

8. Dataset requirements:

a)	Please fill in the following registration form to receive the dataset password: https://forms.gle/3bMHrPDEuPRQ7t7R7
b)	By participating on this competition, you agree to the following Terms and Conditions: https://tuit.cat/nZ1eq