---
license: cc
task_categories:
- text-generation
- text-classification
language:
- en
configs:
- config_name: all
  data_files:
  - path:
    - train.jsonl.zst
    split: train
- config_name: en
  data_files:
  - path:
    - train_en.jsonl.zst
    split: train
  default: true
tags:
- project-gutenberg
- fiction
- literary
- historical
---
# Project Gutenberg Top 100

Project Gutenberg is a library of over 75 000 free eBooks in the public domain.

## Dataset Overview

This dataset contains the 100 most downloaded Project Gutenberg eBooks from May 4, 2025, to May 11, 2025. 
Each eBook was processed to extract clean, paragraph-level text suitable for natural language processing tasks.
<!-- A Portuguese dictionary was omitted. -->

**Key processing steps:**
- Text files were obtained for each eBook.
- Text was split into paragraphs, delimited by double newlines (`\n\n`).
- Markdown formatting was stripped.
- Curly quotes and irregular spacing were normalized.
- Empty paragraphs and paragraphs in all capitals (typically headings or tables of contents) were removed.
- Language detection using langdetect.

## Dataset Structure

There are two configs:
- `en` containing only unique paragraphs detected as English
- `all` containing paragraphs in all languages. Repeated paragraphs were kept.

Each example in the dataset has the following fields:

| Field  | Description                                                   |
|--------|---------------------------------------------------------------|
| file   | Project Gutenberg ID of the eBook (integer)                   |
| index  | Paragraph number within the eBook, starting from 0 (integer)  |
| text   | Paragraph text (string)                                       |
| language | Two-letter language code from automatic detection (string) |

**Example entry:**
```json
{
  "file": 2600,
  "index": 10544,
  "text": "\"The children will live just the same. With such masters one can live.\"",
  "language": "en"
}
```

## Limitations

- A small number of eBooks were not included due to missing text files.
- While the dataset may contain books in multiple languages, the vast majority are in English.
- Paragraphs are not deduplicated to preserve the original narrative flow, including repeated content.
- The language and style reflect the historical, public domain nature of the texts.

## Usage

You can load this dataset with the Hugging Face Datasets library:
```python
from datasets import load_dataset

dataset = load_dataset("agentlans/project-gutenberg-top100", "en")
```
Refer to the [Hugging Face Datasets documentation](https://huggingface.co/docs/datasets/index) for more details on loading and processing datasets.

## Source

- Data collected from: https://www.gutenberg.org/browse/scores/top#books-last7
- Snapshot date: May 11, 2025

## License

Project Gutenberg eBooks are in the public domain in the United States. For more details, see the [Project Gutenberg License](https://www.gutenberg.org/policy/license.html).