Hugging Face's logo

Datasets:
cometadata
/
generic-resource-type-training-data

Modalities:
Text
Formats:
parquet
Size:
100K - 1M
Libraries:
Datasets
Dask
Dataset card Data Studio Files
xet
Community
1
generic-resource-type-training-data / README.md
parthsarin's picture
parthsarin
Update README.md
eed6921 verified
preview code
|
raw
history blame contribute delete
4.38 kB
metadata
dataset_info:
  features:
    - name: description
      dtype: string
    - name: label
      dtype: string
    - name: completion
      dtype: string
  splits:
    - name: train
      num_bytes: 121725759
      num_examples: 242509
  download_size: 35893211
  dataset_size: 121725759
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Generic Resource Type Classification Dataset

This dataset contains training data for classifying academic resources into 32 generic resource types as defined by DataCite metadata standards. The dataset is designed for fine-tuning large language models to improve classification accuracy for the ~25 million works currently classified only with the generic type "Text" in DataCite metadata.

Dataset Description

The dataset is part of the COMET enrichment and curation workflow pilot project focused on improving generic resource type classification. It contains structured metadata descriptions of academic resources along with their corresponding resource type labels.

Dataset Structure

The dataset contains three columns:

  • description: Text description of the resource containing structured metadata fields from DataCite records
  • label: The generic resource type label (e.g., "JournalArticle", "Dataset", "Software")
  • completion: Numeric representation of the resource type (0-31 mapping to the 32 categories)

Resource Type Categories

The dataset classifies resources into 32 categories:

  1. Audiovisual - Visual representations with motion (films, videos)
  2. Award - Funding, grants, scholarships, recognition
  3. Book - Bound collection of pages with text/images
  4. BookChapter - Division of a book
  5. Collection - Aggregation of multiple resources
  6. ComputationalNotebook - Virtual notebook for literate programming
  7. ConferencePaper - Paper intended for conference acceptance
  8. ConferenceProceeding - Collection of conference papers
  9. DataPaper - Publication describing specific datasets
  10. Dataset - Structured data files
  11. Dissertation - Academic thesis (especially PhD)
  12. Event - Time-based occurrences (webcasts, conventions)
  13. Image - Visual representations (photos, drawings)
  14. Instrument - Physical devices for data collection
  15. InteractiveResource - Resources requiring user interaction
  16. Journal - Scholarly periodical publication
  17. JournalArticle - Individual article within a journal
  18. Model - Abstract/mathematical representations
  19. OutputManagementPlan - Research output handling plans
  20. PeerReview - Evaluation by field experts
  21. PhysicalObject - Physical specimens or artifacts
  22. Preprint - Pre-peer-review scholarly papers
  23. Project - Planned collaborative endeavors
  24. Report - Organized information documents
  25. Service - Organized systems for end users
  26. Software - Computer programs and applications
  27. Sound - Audio recordings
  28. Standard - Established reference models
  29. StudyRegistration - Research plan descriptions
  30. Text - General textual resources
  31. Workflow - Structured process sequences
  32. Other - Resources not fitting other categories

Data Source and Processing

The training data is created by:

  1. Sampling from DataCite metadata records
  2. Filtering to exclude generic "Text" and "Other" categories for training
  3. Balancing samples across categories (up to 10,000 examples per category)
  4. Formatting metadata into structured text descriptions

Intended Use

This dataset is designed for:

  • Fine-tuning language models for resource type classification
  • Training models to distinguish between similar resource types (e.g., Software vs Dataset)
  • Improving automated metadata curation for academic repositories
  • Supporting the COMET project's enrichment workflows

Model Training

The dataset is used with:

  • Models: Qwen2.5-7B-Instruct and similar instruction-tuned LLMs
  • Training: LoRA fine-tuning with completion-only loss
  • Evaluation: Accuracy metrics and confusion matrices
  • Inference: Zero-temperature sampling with probability tracking

Limitations

  • Limited to resources with existing DataCite metadata
  • Class imbalance despite sampling efforts (because certain classes are under-represented in the base distribution)
  • Mostly English-language academic resources

Citation

If you use this dataset, please cite the COMET project and DataCite metadata standards.