Semantic Role Labeling Datasets for Crisis Event

This dataset is a collection of Indonesian language text data extracted from Twitter and designed to support research in Semantic Role Labeling (SRL) and Named Entity Recognition (NER) tasks. This dataset offers a combination of argument labels for SRL and entity labels for NER, which are very important in information extraction. Through SRL, the dataset can identify semantic roles, such as who the victim is, where the location is, what the cause is, and how the impact of the crisis event is. Meanwhile, NER can recognize entities such as the name of a place or organization involved.

The data was collected using the Twitter API with relevant keywords from January 2018 to December 2023, covering four major crisis events in Indonesia: fires, accidents, floods, and earthquakes. Two experienced annotators carried out the annotation process—a disaster management expert and a doctoral student in computer science. The level of agreement between annotators was measured using Fleiss' Kappa, with results above 0.80 for all labels, indicating high data quality and consistency in labeling.