Putting Science on the Map: Spatio-Temporal Metadata for Scientific Article Discovery

For tackling the challenge of discovering scientific articles, researchers pursue several options: they conduct general web searches, explore generic academic databases like OpenAlex or Google Scholar, use a discipline-specific portal, task an AI agent, or consider personal recommendations, e.g., via social media. These options typically rely on one of the following approaches: search terms (title, abstracts, keywords, or full text), citation chaining, editorial curation (e.g., topical journals), or known authors and affiliations. However, as the number of publications is continuously rising, there is a need for additional methods that link scientific content in novel ways and help to find relevant works. An underused approach relies on the fact that almost all research has a spatial and temporal component, i.e. the “where” and “when” of a scientific article. How do I find a scientific article by exploiting its geographic context? Currently, if any spatio-temporal metadata can be found at all, it is likely to relate to the author's affiliation or the date of publication rather than the actual content of the research. The latter information is hidden, for example as place names or coordinates, in the full text, supplementary materials, visualisations, or data, but it is not available as human- and machine-readable metadata.

In this work, we present novel tools that are integrating spatio-temporal metadata into the scholarly publishing process: the geoMetadata plugin and the OPTIMAP. The geoMetadata plugin (https://github.com/TIBHannover/geoMetadata) provides authors and journal managers with straightforward tools, such as an interactive map, to collect valid spatio-temporal article metadata during the submission process in the widely used scholarly publishing platform, Open Journal Systems (OJS). The resulting metadata is published in a machine-readable format and articles are made discoverable on maps after publication. Building on this, OPTIMAP (https://github.com/GeoinformationSystems/optimap) demonstrates how scientific articles of several journals can be found via a single map view and published in one open API.

To realise the potential of spatio-temporal metadata fully, a large amount of existing literature needs to be enriched with trustworthy spatio-temporal metadata. We sketch a new framework to support the enrichment of scientific articles in the submission process and for already existing literature. First, various technologies will be evaluated: (i) Named Entity Recognition (NER), that leverage controlled gazetteers to extract place names and temporal expressions (ii) Optical Character Recognition (OCR) to recover spatio-temporal information from maps and figures and (iii) Large Language Models (LLMs) for full-document reasoning. In a second step, the framework will be applied in both an assistance mode (e.g., during the submission process) and a fully automatic mode (back catalogue of journals, publishers, conference series, etc.) for extracting spatio-temporal metadata. The extracted metadata could undergo different curation and validation steps and ultimately become available as part of a discipline-specific knowledge graph or generic academic databases. When such data exists on a large scale, one can explore an extension for scientific search portals, or improvements for handling spatio-temporal metadata throughout the whole research data management (RDM) cycle.