COMET Project: Improve Resource Type Classification in DataCite

The DataCite metadata schema includes a required property resourceTypeGeneral that is used to categorize the types of outputs and resources associated with a DOI. The categorization schema uses a controlled list of 32 types (as of version 4.6) ; some are more general and others are more specific. The DataCite schema also includes an additional property, resourceType, which allows for a free-text description to accompany the controlled value. Nearly 18 million DOI records—20% of DataCite's corpus as of July 2025—carry generic resourceTypeGeneral values. While these categorization choices may be appropriate in some cases; a potential downside is how they might limit precision of discovery and reporting: Text – 16.75% Other – 1.62% Not defined – 1.39% This project explored two automated approaches to reclassify these records into more specific, meaningful categories, improving the precision and utility of DataCite’s metadata for discovery and reporting: Procedural reclassification and model-based classification. Procedural reclassification proved highly effective, successfully reclassifying 3.36 million records where the free-text resourceType field contained information that directly corresponded to a valid resourceTypeGeneral value. Model-based bulk reclassification proved far more challenging than anticipated: records using "Text" or "Other" appear out of distribution from records with specific types, as their metadata appears to either differ significantly from other works or is insufficient as a sole input for reliable classification.