README.md

---
dataset_info:
- config_name: main_group
  features:
  - name: publication_number
    dtype: string
  - name: labels
    sequence: string
  splits:
  - name: train
    num_bytes: 369885895
    num_examples: 7491648
  - name: test
    num_bytes: 43089767
    num_examples: 832405
  download_size: 163987062
  dataset_size: 412975662
- config_name: subgroup
  features:
  - name: publication_number
    dtype: string
  - name: labels
    sequence: string
  splits:
  - name: train
    num_bytes: 693083457
    num_examples: 7492144
  - name: test
    num_bytes: 80785020
    num_examples: 832461
  download_size: 399887593
  dataset_size: 773868477
configs:
- config_name: main_group
  data_files:
  - split: train
    path: main_group/train-*
  - split: test
    path: main_group/test-*
- config_name: subgroup
  data_files:
  - split: train
    path: subgroup/train-*
  - split: test
    path: subgroup/test-*
  default: true
license: cc-by-sa-4.0
task_categories:
- text-classification
tags:
- legal
pretty_name: CPC classification datasets
size_categories:
- 1M<n<10M
---
# CPC classification datasets

These datasets have been used to train the CPC ([Cooperative Patent Classification](https://www.cooperativepatentclassification.org/home)) classification models mentioned in the article **_Hähnke, V. D., Wéry, A., Wirth, M., & Klenner-Bajaja, A. (2025). Encoder models at the European Patent Office: Pre-training and use cases. World Patent Information, 81, 102360. https://doi.org/10.1016/j.wpi.2025.102360_**.

Columns:
- `publication_number`: the patent publication number, the content of the publication can be looked up using e.g. [Espacenet](https://worldwide.espacenet.com/patent/search?q=EP4030126A1) or the [EPO’s Open Patent Services](https://www.epo.org/en/searching-for-patents/data/web-services/ops)
- `labels`: the CPC symbols used as prediction labels (CPC release 2024.01)

## Datasets
### Subgroup dataset
Used to train the _subgroup_ model with 224 542 labels.

How to load the dataset:
```python
from datasets import load_dataset
dataset = load_dataset("mwirth-epo/cpc-classification-data", name="subgroup")
```

### Main group dataset
Used to train the _main group_ model with 9 025 labels.

This dataset was created from the subgroup dataset with a filter excluding main groups with less than 20 documents.

How to load the dataset:
```python
from datasets import load_dataset
dataset = load_dataset("mwirth-epo/cpc-classification-data", name="main_group")
```


## Citation

**BibTeX:**
```bibtex
@article{HAHNKE2025102360,
title = {Encoder models at the European Patent Office: Pre-training and use cases},
journal = {World Patent Information},
volume = {81},
pages = {102360},
year = {2025},
issn = {0172-2190},
doi = {https://doi.org/10.1016/j.wpi.2025.102360},
url = {https://www.sciencedirect.com/science/article/pii/S0172219025000274},
author = {Volker D. Hähnke and Arnaud Wéry and Matthias Wirth and Alexander Klenner-Bajaja},
keywords = {Natural language processing, Language model, Encoder network, Classification, Cooperative Patent Classification}
}
```

**APA:**

Hähnke, V. D., Wéry, A., Wirth, M., & Klenner-Bajaja, A. (2025). Encoder models at the European Patent Office: Pre-training and use cases. World Patent Information, 81, 102360. https://doi.org/10.1016/j.wpi.2025.102360