---
license: mit
task_categories:
- image-to-text
- question-answering
- text-generation
language:
- en
tags:
- chemistry
- biology
- math
- computer-science
- physics
---

# 📚 arXiv Dataset Snapshot (Kaggle → Hugging Face)

This repository hosts a **snapshot of the arXiv dataset**, originally published on **Kaggle**, and re-uploaded to **Hugging Face Datasets** for easier access, versioning, and seamless integration with modern ML workflows.

The goal is to make large-scale arXiv metadata and paper content readily usable for **LLM training, retrieval-augmented generation (RAG), citation analysis, and research analytics**.

---

## 🔍 What’s Inside

Depending on the Kaggle snapshot version, this dataset typically includes:

* 🆔 **arXiv IDs**
* 🏷️ **Titles**
* ✍️ **Authors**
* 🗂️ **Primary & secondary categories**
* 📝 **Abstracts**
* 📅 **Submission / update dates**
* 🔗 **DOI & journal references (when available)**

> ⚠️ This is a *snapshot* — it does **not** auto-update with new arXiv submissions.

---

## 📢 Current Status

* ✅ Metadata is currently available (titles, authors, abstracts, categories, etc.)

* 🚧 Full papers will be uploaded soon, category-wise (e.g., cs, math, physics, stat)

This repository will be updated incrementally as category-wise paper content is added.

---

## 📦 Dataset Structure

A typical layout looks like:

```
.
├── metadata/
│   ├── arxiv-metadata-oai-snapshot.json
├── papers/
│   ├── cs/
│   ├── math/
│   ├── physics/
│   └── ...
├── README.md
```

Or (for lightweight versions):

```
arxiv-metadata-oai-snapshot.json
```

Each record is **one paper per line (JSONL)** for efficient streaming and partial loading.

---

## 🤗 Loading with Hugging Face

```python
from datasets import load_dataset

# Replace with your HF username / org and dataset name
dataset = load_dataset("anuj0456/arxiv-dataset", split="train")

print(dataset[0])
```

### Streaming (recommended for large snapshots)

```python
dataset = load_dataset(
    "anuj0456/arxiv-dataset",
    split="train",
    streaming=True
)

for sample in dataset.take(3):
    print(sample["title"])
```

---

## 🧠 Use Cases

* 📚 **Pretraining / continued training** of language models
* 🔎 **RAG pipelines** (abstracts + full text)
* 📊 **Scientometrics & citation analysis**
* 🧪 **Topic modeling & clustering**
* 🏷️ **Multi-label classification (arXiv categories)**
* 🕸️ **Knowledge graph construction**

---

## ⚖️ License & Attribution

* Original data source: **arXiv.org**
* Kaggle snapshot license: *as provided by Kaggle / arXiv*
* arXiv content is generally distributed under **CC BY 4.0** or **arXiv.org perpetual license**

If you use this dataset, **please cite arXiv appropriately**:

```
@misc{arxiv,
  title = {arXiv e-Print archive},
  url = {https://arxiv.org}
}
```

---

## 🧾 Notes & Limitations

* This dataset may contain:

  * Missing fields
  * Duplicate versions of the same paper
  * LaTeX artifacts in full text
* No guarantees of completeness or correctness
* Not an official arXiv distribution

---

## 🙌 Acknowledgements

* **arXiv.org** for maintaining the open research archive
* **Kaggle** for hosting the dataset snapshot
* **Hugging Face** for dataset infrastructure

---

## 🔗 Related

* arXiv API: [https://arxiv.org/help/api](https://arxiv.org/help/api)
* Kaggle arXiv datasets
* Hugging Face Datasets documentation

---

If you find this dataset useful, consider ⭐ starring the repo or sharing it with the community.

Happy researching! 🚀