Skip to content

EducationalTestingService/sarcasm

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Automatic Sarcasm Detection

The Shared Task (2nd FigLang Workshop at ACL 2020) is now over. Thanks a lot, participants :)

Please refer to reddit and twitter sub-directories for further references on datasets.

For Twitter and Reddit, training and testing datasets are provided for sarcasm detection tasks in jsonlines format.

Each line contains a JSON object with the following fields :

  • label : SARCASM or NOT_SARCASM
  • id: String identifier for sample. This id will be required when making submissions.
    • ONLY in test data
  • response : the sarcastic response, whether a sarcastic Tweet or a Reddit post
  • 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, then c2 is a reply to c1 and c3 is a reply to c2. Further, if the sarcastic response is r, then r is a reply to c3.

For instance, for the following training example :

"label": "SARCASM", "response": "Did Kelly just call someone else messy? Baaaahaaahahahaha", "context": ["X is looking a First Lady should . #classact, "didn't think it was tailored enough it looked messy"]

The response tweet, "Did Kelly..." is a reply to its immediate context "didn't think it was tailored..." which is a reply to "X is looking...". Your goal is to predict the label of the "response" while also using the context (i.e, the immediate or the full context).

Dataset size statistics :

Train Test
Reddit 4400 1800
Twitter 5000 1800

For Test, we will be providing you the response and the context. We will also provide the id (i.e., identifier) to report the the results.

Submission Instructions : Please follow the given link

Main References:

A Report on the 2020 Sarcasm Detection Shared Task. Debanjan Ghosh, Avijit Vajpyee, Smaranda Muresan. Proceedings of the Second Workshop on Figurative Language Processing.


Note: Since we have collected our training data from popular social media platforms a large portion of the utterances are on controversial and/or political and social topics. Although we have pre-processed the training data and lightly edited to remove contentious text, many utterances still contain controversial perspectives (of the users) and informal language.