This repository contains our efforts to detect tweets calling for help, specifically after the 2023 Turkiye-Syria Earthquake.
The video demonstration of our tool is given at demo_video.zip. We remove any information that may reveal the personal identity in the video to ensure privacy.
The tweet IDs (1,000) that are used in the demonstration video.
The annotated dataset is given at dataset.tsv. We annotate 1,000 tweets in Turkish if tweets call for help (i.e. request rescue, supply, or donation), and their entity tags (person, city, address, status).
Column Name | Description |
---|---|
label | Human annotation if tweet calls for help (binary classification) |
entities | Human annotation of entity tags (i.e. person, city, address, and status). The format is [START_INDEX]:[END_INDEX]%[TAG_TYPE]. |
tweet_id | Tweet ID from Twitter API. |
Distibution of tweets in the dataset is as follows:
Label | Tweet Count | ADDRESS | CITY | PERSON | STATUS | ||||
---|---|---|---|---|---|---|---|---|---|
Call for Help | 418 | 516 | 504 | 364 | 392 | ||||
Not Call for Help | 582 | - | - | - | - | ||||
Total | 1,000 tweets | 1,776 tags |
The keyword lists that we use to collect tweets are given in general_keywords.txt and help_keywords.txt.
The deployed models are published at https://huggingface.co/ctoraman
If you make use of this dataset, please cite following paper.
@misc{toraman2023earthquake,
doi = {10.48550/ARXIV.2302.13403},
url = {https://arxiv.org/abs/2302.13403},
author = {Toraman, Cagri and Kucukkaya, Izzet Emre and Ozcelik, Oguzhan and Sahin, Umitcan},
keywords = {Social and Information Networks (cs.SI), Computation and Language (cs.CL), Information Retrieval (cs.IR), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {Tweets Under the Rubble: Detection of Messages Calling for Help in Earthquake Disaster},
publisher = {arXiv},
year = {2023},
copyright = {Creative Commons Attribution Non Commercial Share Alike 4.0 International}
}