CoolFace
Datasetpublic

academic-paper-surgical-edit/academic-flawed-papers-dataset

🧪 Surgical Flaws Dataset This dataset is a companion resource to the main structured-paper collection: 👉 https://huggingface.co/datasets/academic-paper-surgical-edit/academic-papers-dataset While the original dataset contains clean, parsed scientific papers,this dataset introduces controlled, surgically injected flaws into those papers. 📝 Dataset Summary This dataset augments selected papers from: ICLR → 2024, 2025 ICML → 2024, 2025 NeurIPS → 2021, 2022… See the full description on the dataset page: https://huggingface.co/datasets/academic-paper-surgical-edit/academic-flawed-papers-dataset.

sourceHugging Facecc-by-4.0updated 8mo agoView on Hugging Face
0likes17downloads
Dataset Card

🧪 Surgical Flaws Dataset

This dataset is a companion resource to the main structured-paper collection:

👉 https://huggingface.co/datasets/academic-paper-surgical-edit/academic-papers-dataset

While the original dataset contains clean, parsed scientific papers, this dataset introduces controlled, surgically injected flaws into those papers.

📝 Dataset Summary

This dataset augments selected papers from:

  • —ICLR → 2024, 2025
  • —ICML → 2024, 2025
  • —NeurIPS → 2021, 2022, 2023, 2024

All flaws are injected with controlled using LLM-guided rewriting tools.

📊 Dataset Statistics

Conference Paper Counts

(These numbers reflect the subset used for flaw injection; not the full conference totals.)

[image]

ConferenceAcceptedRejected
NeurIPS 20241036—
NeurIPS 2023781—
NeurIPS 202280238
NeurIPS 202189524
ICML 202544122
ICML 20244—
ICLR 20251000797
ICLR 2024674518

🌍 Languages

  • —English (en) (Official conference submissions from ICLR, ICML, NeurIPS)

🚀 How to Use This Dataset

Because the dataset contains large ZIP archives and thousands of structured papers, we recommend using git-xet for efficient cloning and storage.

1️⃣ Install git-xet

git-xet enables fast and storage-efficient checkout of large files from Hugging Face Datasets.

Official docs: https://hf.co/docs/hub/git-xet

If you're on macOS:

bash
brew install git-xet
git xet install

Linux:

bash
curl https://gitxet.com/install.sh | sudo bash
git xet install

2️⃣ Clone the dataset repository

bash
git clone https://huggingface.co/datasets/academic-paper-surgical-edit/academic-flawed-papers-dataset

This will download lightweight file pointers first, and git-xet will fetch the large paper archives on demand.

📂 Dataset Structure

Inside the cloned repo, you will find per-conference ZIP files:

To extract:

bash
unzip {ConferenceName}.zip -d {ConferenceName}/

After extraction, you will see:

content/
└── output/
    └── {ConferenceName}_latest_flawed_papers_v1/
        └── flawed_papers_global_summary.csv
        └── accepted/
            └── {ID}/
                └── {ID}_modifications_summary.csv
                └── flawed_papers/
                    │   ├── insufficient_random_seeds.md
                    │   └── ...
        └── rejected/
            └── {ID}/
                └── {ID}_modifications_summary.csv
                └── flawed_papers/
                    │   ├── insufficient_random_seeds.md
                    │   └── ...

File Descriptions

File / FolderPurpose
{ID}_modifications_summary.csvProvides a per-paper summary of injected flaws, including the flaw ID and description.
flawed_papers/Contains the manipulated paper files, each altered according to a specific injected flaw.
flawed_papers_global_summary.csvAggregated summary listing every paper–flaw pair, including the global flaw ID, flaw category, and description.

📜 Citation

TBA