---
license: cc-by-4.0
task_categories:
  - text-generation
  - feature-extraction
language:
  - en
  - multilingual
tags:
  - science
  - papers
  - fulltext
  - scientific-papers
size_categories:
  - 10M<n<100M
configs:
  - config_name: corex
    data_files: "corex/*.parquet"
  - config_name: papers-2
    data_files: "papers-2/*.parquet"
  - config_name: papers-3
    data_files: "papers-3/*.parquet"
  - config_name: pes2o
    data_files: "pes2o/*.parquet"
---

# Scientific Papers - Raw Full Text

**~57 million scientific papers with full text**, extracted from multiple large-scale academic paper collections. This dataset provides raw full text suitable for pre-training, fine-tuning, or building search indices over scientific literature.

## Subsets

| Subset | Papers | Size | Source |
|--------|--------|------|--------|
| **papers-2** | ~18.5M | ~358 GB | S2ORC papers collection (untitled subset) |
| **papers-3** | ~27.4M | ~198 GB | S2ORC scientific-papers collection |
| **pes2o** | ~8.2M | ~106 GB | Pes2oX-fulltext (Semantic Scholar, Apache 2.0) |
| **corex** | ~2.9M | ~25 GB | CORE repository (core.ac.uk, Apache 2.0) |
| **Total** | **~57M** | **~687 GB** | |

## Schema (9 columns)

| Column | Type | Description |
|--------|------|-------------|
| `paper_id` | string | Unique identifier (source-specific) |
| `title` | string | Paper title (when available) |
| `authors` | string | Author names (semicolon-separated) |
| `year` | string | Publication year (when available) |
| `venue` | string | Publication venue (when available) |
| `doi` | string | DOI (when available) |
| `abstract` | string | Paper abstract (when available) |
| `raw_fulltext` | string | Complete paper text |
| `text_length` | int64 | Character count of full text |

## Usage

```python
from datasets import load_dataset

# Load a specific subset
ds = load_dataset("scientifi-papers/scientific-papers", "corex")

# Load papers-3 (largest)
ds = load_dataset("scientifi-papers/scientific-papers", "papers-3")

# Access full text
paper = ds['train'][0]
print(f"Title: {paper['title']}")
print(f"Text length: {paper['text_length']} chars")
print(paper['raw_fulltext'][:500])
```

## Sources

- **papers-2 / papers-3**: Extracted from the Semantic Scholar Open Research Corpus (S2ORC) via GROBID PDF parsing
- **pes2o**: Allen AI's Pes2oX-fulltext dataset (cleaned S2ORC full text)
- **corex**: CORE repository (core.ac.uk) - 2018 full-text dump of open-access research papers

## License

CC-BY-4.0