Task 1: Offline Harm Potential Identification (HarmPot-ID)
TRAC-2024 will introduce the novel task of predicting the offline harm potential of social media posts - broadly the task is to predict whether a specific post is likely to initiate, incite or further exaggerate an offline harm event (viz. riots, mob lynching, murder, rape, etc). It will consist of two sub-tasks.
Sub-task 1a: What is the offline harm potential of a document?
This will be a four-class classification task where the participants will be required to predict the level of offline harm potential -
0 (it will never lead to offline harm, in any context),
1 (it could lead to incite an offline harm event given specific conditions or context),
2 (it is most likely to incite in most contexts or probably initiate an offline harm event in specific contexts)
3 (it is certainly going to incite or initiate an offline harm event in any context).
Sub-task 1b: Who is/are the most likely target(s) of the offline harm?
If an offline harm event is triggered, who are going to be the most affected groups of people? In this task, only the broad category of identities of the target(s) are to be predicted. It will be a five-class classification task -
Gender
Religion
Descent
Caste
Political Ideology