Identifying and extracting Data Access Statements from full-text academic articles

TL;DR

This paper presents a machine learning-based module developed by CORE to automatically identify and extract Data Access Statements from full-text academic articles, enhancing metadata quality and data sharing practices.

Contribution

The paper introduces a novel machine learning tool for automated DAS identification and extraction, improving efficiency and standardization in scholarly metadata curation.

Findings

Reduces manual curation workload

Improves metadata accuracy and completeness

Supports compliance with funder data-sharing policies

Abstract

A Data Access Statement (DAS) is a formal declaration detailing how and where the underlying research data associated with a publication can be accessed. It promotes transparency, reproducibility, and compliance with funder and publisher data-sharing requirements. Funders such as Plan S, the European Union, UKRI, and NIH emphasise the inclusion of DAS in publications, underscoring its growing importance. While a DAS enhances research by increasing transparency, discoverability, and data quality while clarifying access protocols and elevating datasets as first-class research outputs, the repository community faces challenges in managing and curating DAS as a standard metadata component. Manual DAS curation remains labour-intensive and time-consuming, hindering efficient data-sharing practices. CORE has co-designed with the repository community a module that uses machine learning to identify and extract DAS from full-text articles. This tool facilitates the automated encoding, curation, and validation of DAS within metadata, reducing manual workload and improving metadata quality. This integration aligns with CORE's objective to enhance repository services by providing enriched metadata and supporting compliance with funder requirements. By streamlining DAS management and expanding metadata frameworks, CORE contributes to a more accessible and interconnected scholarly ecosystem, fostering data discoverability and reuse.