A transparent knowledge-based framework for large-scale gender inference using the World Gender Name Dictionary

Name-based gender inference is often used when self-reported gender is not available, but many tools give little indication of why a particular label was assigned. In this article we develop a knowledge-based approach centred on the World Gender Name Dictionary (WGND). The method combines country, language, and global name evidence, reconstructs weighted language and global dictionaries from the original country-level records, and uses confidence thresholds to leave uncertain cases unresolved. We also test a small set of explicit cultural rules based on Vietnamese naming elements, Arabic/Islamic patronymics, and Slavic surname morphology. The same 330,994 ORCID authors served to develop the pipeline and assess its results. Against these biography-derived reference labels, the WGND pipeline reached 95.78% precision with 89.67% coverage. Most assignments came from the first, high-confidence Global stage; the Country and Language stages were used for authors still without a label. Cultural markers recovered relatively few additional cases, at 97.9% precision. Remaining disagreements mainly concerned names shared across cultures, unisex forms, transliterations and abbreviations. The proposed framework provides an auditable alternative to black-box gender inference and illustrates how heterogeneous evidence sources can be integrated within scalable knowledge-based systems. Preprint status: This manuscript has been submitted to Knowledge-Based Systems and is currently undergoing editorial evaluation. It has not yet been peer reviewed or accepted for publication. Its content may change during the review process.