Skip to content

Dataset · ccdv

Patent texts and abstracts for classification

ccdv/patent-classification

35k English patents and abstracts divided into 9 classes for long-context evaluation.

What's inside

An English-language dataset built upon the well-known BIGPATENT corpus. It includes 35k documents (full patent texts and abstracts) split across 3 partitions and 9 imbalanced categories. Because patent texts without abstracts exceed 512 tokens, the dataset is specifically designed for long-document classification tasks.

What you can do with it

  • Training and evaluating classification models on long texts exceeding 512 tokens
  • Automated categorization of patent and technical documentation
  • Testing classifier robustness against imbalanced classes
Tasks
Text classification
Languages
English
Size
10K–100K rows
Downloads this month
971
Likes
31

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("ccdv/patent-classification")