README.md

metadata

dataset_info:
  features:
    - name: text
      dtype: string
    - name: labels
      sequence:
        class_label:
          names:
            '0': joy
            '1': trust
            '2': fear
            '3': surprise
            '4': sadness
            '5': neutral
            '6': anticipation
            '7': anger
            '8': disgust
    - name: id
      dtype: string
  splits:
    - name: train
      num_bytes: 4224138
      num_examples: 43410
    - name: validation
      num_bytes: 527119
      num_examples: 5426
    - name: test
      num_bytes: 524443
      num_examples: 5427
  download_size: 3427263
  dataset_size: 5275700
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
task_categories:
  - text-classification
language:
  - en
tags:
  - nlp

Original dataset: GoEmotions dataset

Filtered using the following mapping based on the basic emotions found in Plutchik's Wheel of Emotions

wheel_dict = {
    "joy": [
        "joy",
        "amusement",
        "excitement",
        "gratitude",
        "pride",
        "relief",
        "admiration",
        "love",
        "optimism",
    ],
    "trust": ["approval", "caring"],
    "fear": ["fear", "nervousness"],
    "surprise": ["surprise", "realization", "confusion", "curiosity"],
    "sadness": ["sadness", "disappointment", "grief", "remorse", "embarrassment"],
    "neutral": ["neutral"],
    "anticipation": ["desire"],
    "anger": ["anger", "annoyance", "disapproval"],
    "disgust": ["disgust"],
}

Dataset contains 7 emotion labels as per Dr. Ekman theory. Labels are as follows:

  '0': joy
  '1': trust
  '2': fear
  '3': surprise
  '4': sadness
  '5': neutral
  '6': anticipation
  '7': anger
  '8': disgust