Dataset format:
Each line contains a JSON object with the following fields :
-
response : the Tweet to be classified
-
context : the conversation context of the response
- Note, the context is an ordered list of dialogue, i.e., if the context contains three elements,
c1
,c2
,c3
, in that order, thenc2
is a reply toc1
andc3
is a reply toc2
. Further, the Tweet to be classified is a reply toc3
.
- Note, the context is an ordered list of dialogue, i.e., if the context contains three elements,
-
label :
SARCASM
orNOT_SARCASM
-
id: String identifier for sample. This id will be required when making submissions. (ONLY in test data)
For instance, for the following training example :
"label": "SARCASM", "response": "@USER @USER @USER I don't get this .. obviously you do care or you would've moved right along .. instead you decided to care and troll her ..", "context": ["A minor child deserves privacy and should be kept out of politics . Pamela Karlan , you should be ashamed of your very angry and obviously biased public pandering , and using a child to do it .", "@USER If your child isn't named Barron ... #BeBest Melania couldn't care less . Fact . ๐ฏ"]
The response tweet, "@USER @USER @USER I don't get this..." is a reply to its immediate context "@USER If your child isn't..." which is a reply to "A minor child deserves privacy...". Your goal is to predict the label of the "response" while optionally using the context (i.e, the immediate or the full context).
Dataset size statistics :
Train | Test |
---|---|
5000 | 1800 |
For Test, we've provided you the response and the context. We also provide the id (i.e., identifier) to report the results.
Submission Instructions : Please add a comma separated file named answer.txt
containing the predictions on the test dataset. The file should have no headers and have exactly 1800 rows. Each row must have the sample id and the predicted label. For example:
twitter_1,SARCASM
twitter_2,NOT_SARCASM
...