---
dataset_info:
- config_name: cc-by
  features:
  - name: id
    dtype: string
  - name: idx
    dtype: int64
  - name: paragraph
    dtype: string
  splits:
  - name: train
    num_bytes: 151115318
    num_examples: 139463
  download_size: 76199216
  dataset_size: 151115318
- config_name: cc-by-nc
  features:
  - name: id
    dtype: string
  - name: idx
    dtype: int64
  - name: paragraph
    dtype: string
  splits:
  - name: train
    num_bytes: 78538396
    num_examples: 69457
  download_size: 39741294
  dataset_size: 78538396
configs:
- config_name: cc-by
  data_files:
  - split: train
    path: cc-by/train-*
- config_name: cc-by-nc
  data_files:
  - split: train
    path: cc-by-nc/train-*
---
# ChemRxiv Paragraphs

This dataset consists of paragraphs from ChemRxiv papers with **CC BY 4.0** and **CC BY-NC 4.0** licenses, sourced from the [BASF-AI/ChemRxiv-Papers](https://huggingface.co/datasets/BASF-AI/ChemRxiv-Papers) dataset. Paragraphs are extracted using [Grobid](https://github.com/kermitt2/grobid), and filtered using an average log word probability method similar to the approach in [allenai/peS2o](https://huggingface.co/datasets/allenai/peS2o). Paragraphs with fewer than 50 words are excluded.

The number of unique papers in each license category is as follows:

- **CC BY 4.0:** 5,848 papers
- **CC BY-NC 4.0:** 3,082 papers

To obtain metadata for each paper, join on the `id` column with the [BASF-AI/ChemRxiv-Papers](https://huggingface.co/datasets/BASF-AI/ChemRxiv-Papers) dataset.

To access paragraphs for a specific license, use the `name` argument as follows:


```python
import datasets

cc_by = datasets.load_dataset('BASF-AI/ChemRxiv-Paragraphs', name='cc-by')
cc_by_nc = datasets.load_dataset('BASF-AI/ChemRxiv-Paragraphs', name='cc-by-nc')

```