HF-NLP10K: A Dataset and Metadata Analysis of 10,000+ Hugging Face NLP Models

This repository contains the data, scripts, and figures accompanying the paper:

HF-NLP10K: A Dataset and Metadata Analysis of 10,000+ Hugging Face NLP Models
(Under submission)

---

🌍 Overview

Hugging Face has become a central platform for sharing NLP models. However, inconsistent and incomplete metadata make it difficult to find suitable models.
HF-NLP10K introduces a structured dataset of over 10,000 NLP models enriched with 33 metadata fields, combining Hugging Face API data with LLM-based parsing of free-text model cards.

---

📦 Repository Structure

data/
├── raw/                 # Raw extracted data from Hugging Face API
│   ├── Data-Stats.csv.gz
│   └── S1-All_NLP.csv.gz
└── final/               # Final structured dataset
    └── HLT-NLP10K.csv

src/
├── preprocessing/       # Scripts for dataset construction (S1–S8)
├── analysis/            # Exploratory data analysis scripts
└── figures/             # Figure generation scripts (for paper visuals)

outputs/
└── figures/             # Generated figures and plots

requirements.txt          # Python dependencies
README.md                 # Overview and usage instructions

---

🧪 Dataset Highlights

• Covers 10,022 NLP models with at least 1,000 downloads
• Includes 33 metadata fields, such as:
  • Pipeline tag, license, training data, evaluation metrics, base model
  • Download counts, fine-tuning compatibility, model card sections (limitations, use cases, etc.)
• Combines API enrichment with LLM-based model card parsing

---

🔧 Reproducing the Dataset

Run the scripts sequentially from S1 to S8:

cd src/preprocessing

python S1-All_NLP_Models.py
python S2-Filter_NLP_Models.py
...
python S8-UniqueRows_FinalDataset.py

The final dataset (HLT-NLP10K.csv) will be available under data/final/.

---

📊 Figures

The figures presented in the paper can be reproduced using:

cd src/figures
python Figure1_NLPGrowth.py
python Figure2_No.ofDownloads.py
...
python Figure5_MissingSections.py

Resulting images will be stored under outputs/figures/.

---

⚙️ Requirements

Install dependencies before running the scripts:

pip install -r requirements.txt

---

🧩 Citation

If you use this dataset or scripts, please cite:

HF-NLP10K: A Dataset and Metadata Analysis of 10,000+ Hugging Face NLP Models. Under submission.

---