Data Gatherer

Data Gatherer is a Python library for automatically extracting dataset references from scientific publications. It processes full-text articles—whether in HTML or XML format—and uses both rule-based and LLM-based methods to identify and structure dataset citations.

What It Does

• Parses scientific articles from open-access sources like PubMed Central (PMC).
• Extracts dataset mentions from structured sections (e.g., Data Availability, Supplementary Material).
• Supports two main strategies:
  • Retrieve-Then-Read (RTR): First retrieves relevant sections using hand-crafted rules, then applies LLMs.
  • Full-Document Read (FDR): Applies LLMs to the full text without section filtering.
• Outputs structured results in JSON format.
• Includes support for known repositories (e.g., GEO, PRIDE, MassIVE) via a configurable ontology.

Use Cases

• Helping data curators and librarians identify datasets cited in publications.
• Supporting meta-analysis and secondary data discovery.
• Enabling dataset indexing and retrieval across the open-access literature.

Citation

If you use Data Gatherer in your research or project, please cite our paper:

@inproceedings{marini_data_2025,
        title = {Data {Gatherer}: {LLM}-{Powered} {Dataset} {Reference} {Extraction} from {Scientific} {Literature}},
        url = {https://aclanthology.org/2025.sdp-1.10},
        doi = {10.18653/v1/2025.sdp-1.10},
        booktitle = {Proceedings of the {Fifth} {Workshop} on {Scholarly} {Document} {Processing} ({SDP} 2025)},
        publisher = {Association for Computational Linguistics},
        author = {Marini, Pietro and Santos, Aécio and Contaxis, Nicole and Freire, Juliana},
        year = {2025},
        pages = {114--123},
}