README.md

metadata

language:
  - en
license: cc-by-4.0
dataset_info:
  features:
    - name: text
      dtype: string
    - name: label
      dtype: string
  splits:
    - name: train
      num_bytes: 10012894
      num_examples: 10000
    - name: test
      num_bytes: 10232963
      num_examples: 10000
  download_size: 11473381
  dataset_size: 20245857
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*

arXiv Abstract Label 20k

This dataset contains abstracts and primary categories from arXiv papers.

Dataset Structure

The dataset is split into two parts: train and test. Each entry contains the following fields:

• text: The abstract of the arXiv paper.
• label: The primary category of the arXiv paper.

The dataset consists of a total of 20,000 entries, with 10,000 for training and 10,000 for testing. Each primary category includes 1,250 entries per split, making the dataset balanced across all categories.

Categories

The dataset includes the following primary categories:

• Computer Science (cs)
• Economics (econ)
• Electrical Engineering and Systems Science (eess)
• Mathematics (math)
• Physics (physics)
• Quantitative Biology (q-bio)
• Quantitative Finance (q-fin)
• Statistics (stat)

Dataset Creation

This dataset was created by using the arxiv library to query arXiv for papers across multiple categories. We fetched the abstracts and primary categories of these papers, sorted by relevance, and organized them into the Hugging Face dataset format.

Usage

You can load the dataset using the datasets library in Python:

from datasets import load_dataset

dataset = load_dataset('Voice49/arXiv-Abstract-Label-20k')