TimSchopf/arxiv_categories

📄 Paper: Efficient Few-shot Learning for Multi-label Classification of Scientific Documents with Many Classes (ICNLSP 2024)

💻 GitHub: https://github.com/sebischair/FusionSent

This is a dataset of scientific documents derived from arXiv metadata. The arXiv metadata provides information about more than 2 million scholarly articles published in arXiv from various scientific fields. We use this metadata to create a dataset of 203,961 titles and abstracts categorized into 130 different classes. To this end, we first perform stratified downsampling of the metadata to only 10% of all articles while retaining the original class distribution. Afterward, articles assigned to categories occurring less than 100 times in the downsampled dataset are removed. To obtain the final dataset, we then perform a stratified train/validation/test split of the processed dataset in an 80:10:10 ratio. The number of examples in each set is shown in the table below.

• The default subset contains the dataset with the document categories as classes in the form of lists of strings. The categories are ordered hierarchically according to the arXiv category taxonomy. In this dataset, the -> symbols indicate a parent->child relationship between categories that can be linked and create a path from the root to the leaf node. For classification, you can either use the complete paths as classes or just parse the respective leaf nodes as classes, resulting in the same (abbreviated) categories.
• The arxiv_category_descriptions subset contains the tags, names, and textual descriptions of the leaf nodes from the arXiv category taxonomy.

Split	Number of Samples
Train	163,168
Validation	20,396
Test	20,397

Each article in the resulting arXiv dataset is categorized into one or more distinct categories. The figure below shows the distribution of papers across the 130 categories of the dataset.

License

MIT

Citation information

When citing our work in academic papers and theses, please use this BibTeX entry:

@inproceedings{schopf-etal-2024-efficient,
    title = "Efficient Few-shot Learning for Multi-label Classification of Scientific Documents with Many Classes",
    author = "Schopf, Tim  and
      Blatzheim, Alexander  and
      Machner, Nektarios  and
      Matthes, Florian",
    editor = "Abbas, Mourad  and
      Freihat, Abed Alhakim",
    booktitle = "Proceedings of the 7th International Conference on Natural Language and Speech Processing (ICNLSP 2024)",
    month = oct,
    year = "2024",
    address = "Trento",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.icnlsp-1.21",
    pages = "186--198",
}