Skip to content

Dataset · CogComp

English Question Type Classification (TREC)

CogComp/trec

6,000 manually labeled English questions for training intent and answer type classifiers.

What's inside

The classic TREC conference dataset contains 5,500 training and 500 testing English questions gathered from four sources and manually labeled. Each question belongs to one of 6 broad and 50 fine-grained categories. The average sentence length is 10 words, with a vocabulary size of 8,700 words.

What you can do with it

  • Training question type classifiers for search and QA systems
  • Benchmarking lightweight text models and embeddings on text-classification tasks
Tasks
Text classification
Languages
English
Size
1K–10K rows
Downloads this month
14.7K
Likes
49

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("CogComp/trec")