What dataset should I use for text classification?

It depends on your goal, but here are some good default choices for text classification datasets:

Good starter datasets

• IMDb Reviews — binary sentiment classification, clean and widely used.
• AG News — 4-way topic classification, simple and fast to benchmark.
• DBpedia — multi-class topic classification with structured labels.
• SST-2 — sentence-level sentiment, common for quick experiments.

If you want a more realistic/challenging task

• Amazon Reviews — large-scale sentiment classification.
• 20 Newsgroups — classic multi-class text classification.
• Yelp Reviews — sentiment and rating prediction.
• Toxic Comment / Jigsaw — multi-label toxicity detection.

If you need a domain-specific dataset

• PubMed / medical abstracts for biomedical classification
• Reuters-21578 for news categorization
• Enron emails for spam/ham or email categorization
• Legal or finance datasets if your use case is specialized

My recommendation

If you’re just starting:

• Use AG News for topic classification
• Use IMDb for sentiment classification

If you want, I can also suggest a dataset based on:

1. binary vs multi-class vs multi-label
2. small vs large dataset
3. Python/Hugging Face availability
4. your domain