3.7 Relationship Metadata
Relationship metadata captures the associations between datasets and related digital assets, including additional datasets, analytical scripts, publications, and derivative data products. To facilitate thorough navigation and improved resource discovery, expressing these relationships bidirectionally is advisable whenever feasible. The adoption of persistent identifiers (such as DOIs or handles) is highly recommended to ensure reliable and long-term connections. Additionally, leveraging artificial intelligence for automated extraction and maintenance of these links can further enhance efficiency and accuracy.
There are several main types of relationships to consider:
Between datasets – For instance, a survey might be the source for an indicator, or an anonymized version of a dataset could be derived from raw data.[1] While current metadata standards allow documentation of such links, they are often recorded in only one direction. Keeping such connections updated takes effort and isn't usually prioritized.
Between data and publications – Metadata should show which datasets are cited in publications and which publications mention specific datasets. The Office of the Chief Statistician develops and maintains an AI-based solution to automatically extract data citations, helping build a structured and up-to-date database linking datasets and World Bank publications. The use of persistent identifiers like DOIs greatly improves precision, especially given inconsistent citation practices.
Between datasets, scripts, and analytical outputs – Scripts used for data processing and analysis connect datasets to the results they produce. Metadata about these relationships can link scripts, datasets, and outputs like tables, figures, visualizations, and reports. The metadata standard used to document data processing and analysis scripts provides the elements needed to establish such relationships.[2]
Creating and maintaining metadata that captures relationships between data resources is resource-intensive and time-consuming. Given ongoing needs to strengthen core metadata—descriptive and structural—by improving completeness, accuracy, quality control, and alignment with standards, relationship metadata is not currently prioritized.
In time series, where each value may have a different source, the identification of the source may be included as a text variable in the time series database instead of being provided in the metadata. ↩︎
For an example of implementation, see the Bank’s reproducibility platform at http://reproducibility.worldbank.org ↩︎