README.md

metadata

dataset_info:
  features:
    - name: text
      dtype: large_string
    - name: hate_speech_score
      dtype: float64
    - name: label
      dtype: int64
  splits:
    - name: train
      num_bytes: 23823783
      num_examples: 135556
  download_size: 10181817
  dataset_size: 23823783
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
task_categories:
  - text-classification
language:
  - en
pretty_name: UCBerkley DLab Measuring hate speech in 3 classes
size_categories:
  - 100K<n<1M

This is modified version of UCBerkley DLab dataset, converted to represtent harmfull level in 3 classes:

1. SAFE - hate_speech_score < -1
2. BORDERLINE - -1 < hate_speech_score < 0.5
3. HARMFUL - hate_speech_score >= 0.5

Text and harmful level classification was created by UCBerkley DLab Team! I'm not creator of this data - i'm only converted it into classes!