shifaa-processed / README.md
noone-m's picture
Update README.md
dbbff95 verified
|
Raw History Blame Contribute Delete
1.27 kB
metadata
dataset_info:
  features:
    - name: Question
      dtype: string
    - name: Main Category
      dtype: string
  splits:
    - name: train
      num_bytes: 39980973
      num_examples: 38912
    - name: validation
      num_bytes: 4434743
      num_examples: 4324
    - name: test
      num_bytes: 4936269
      num_examples: 4805
  download_size: 23971266
  dataset_size: 49351985
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
tags:
  - medical
  - arabic
  - healthcare
pretty_name: 'Shifaa Arabic Medical Consultations - Processed '
size_categories:
  - 10K<n<100K

Shifaa Arabic Medical Consultations - Processed for Classification

Processed version of Ahmed-Selem/Shifaa_Arabic_Medical_Consultations.

Cleaning Steps

  • Extracted main category from hierarchical labels.
  • Removed questions with conflicting categories (for clean single-label supervision).
  • Deduplicated questions.
  • Stratified split: ~81% train, 9% val, 10% test.

Task

Single-label classification: Predict the main medical specialty from the patient's question.