melisekm's picture
Update README.md
b16966b verified
metadata
task_categories:
  - text-classification
language:
  - en
tags:
  - natural disasters
  - tweets
  - classification
  - catastrophic events
pretty_name: Natural Disasters from Social Media
size_categories:
  - 100K<n<1M
annotations_creators:
  - crowdsourced
  - expert-generated
source_datasets:
  - Kaggle 1 - jannesklaas/disasters-on-social-media
  - Kaggle 2 - vstepanenko/disaster-tweets
  - Kaggle 3 - sidharth178/disaster-response-messages
  - 'Zahra et al. - doi: 10.1016/j.ipm.2019.102107'
  - 'CrisisMMD - arxiv: 1805.00713'
  - 'Alam et al. - arxiv: 1805.05151'
  - 'CrisisLexT26 - doi: 10.1145/2675133.2675242'
  - 'Imran et al. - aclanthology: L16-1259'
  - 'CrisisLexT6 - doi: 10.1609/icwsm.v8i1.14538'
  - 'HumAID - doi: 10.1609/icwsm.v15i1.18116'
  - 'CrisisBench - doi: 10.1609/icwsm.v15i1.18115'
configs:
  - config_name: default
    default: true
    data_files:
      - split: train
        path: train.csv
      - split: validation
        path: validation.csv
      - split: test
        path: test.csv
  - config_name: full
    data_files: meta/natural-disasters-from-social-media.csv
  - config_name: meta
    data_files: meta/distributions/*.csv
dataset_info:
  config_name: default
  splits:
    - name: train
      num_bytes: 39817704
      num_examples: 169109
    - name: validation
      num_bytes: 4977163
      num_examples: 21139
    - name: test
      num_bytes: 4981112
      num_examples: 21139
  dataset_size: 49775824

Description

Dataset created for Master's thesis "Detection of Catastrophic Events from Social Media" at the Slovak Technical University Faculty of Informatics.

Contains posts from social media that are split into two categories:

  • Informative - related and informative in regards to natural disasters
  • Non-Informative - unrelated to natural disasters

Other metadata include event type, source dataset etc. To balance classes, 50k tweets from twitter archive for years 2017-2022 were added.

Distributions

Distributions

Source Datasets:

Name Count
Kaggle 1 - URL 951
Kaggle 2 - URL 579
Kaggle 3 - URL 3782
Zahra et al. - URL 6494
CrisisMMD - URL 11043
Alam et al. - URL 11133
CrisisLexT26 - URL 14998
Imran et al. - URL 16549
CrisisLexT6 - URL 22672
HumAID - URL 42837
CrisisBench - URL 31158
ArchiveTeam - URL 49191
Total 211387

Total Event counts:

Type Non-Informative Informative Total
Unknown 61880 14740 76620
Storm 20944 47301 68245
Flood 13104 14637 27741
Earthquake 7844 15549 23393
Fire 2343 8595 10938
Landslide 2392 384 2776
Meteorite 193 545 738
Haze 51 503 554
Volcano 243 139 382