README.md

metadata

dataset_info:
  features:
    - name: text
      dtype: string
    - name: category
      dtype: string
    - name: labels
      dtype: int64
  splits:
    - name: train
      num_bytes: 80088185
      num_examples: 36000
    - name: test
      num_bytes: 1802902
      num_examples: 1200
  download_size: 46779515
  dataset_size: 81891087
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
task_categories:
  - text-classification
  - text-generation
language:
  - tr
size_categories:
  - 100K<n<1M

Dataset Summary

This is a diverse Turkish corpus consisting of 120,000 rows, categorized into 5 major domains.

Dataset Schema & Labels

Category	Labels	Description
Conversation	0	Daily dialogues and conversational language
Review	1	User reviews and sentiment-related data
Wiki	2	Encyclopedic and formal informational texts
News	3	News articles and journalistic language
STEM	4	Academic texts (Science, Tech, Engineering, Math, Medicine)

---

Sources

The dataset is curated from the following sources:

• Conversation: TFLai/Turkish-Dialog-Dataset
• Review: winvoker/turkish-sentiment-analysis-dataset
• Wiki: wikimedia/wikipedia (Turkish)
• News: batubayk/TR-News
• STEM: * hazal/Turkish-Biomedical-corpus-trM
  • hazal/electronic-radiology-phd-thesis-trR
  • ysdede/khanacademy-turkish