Deduplicated OpenAlex-Derived Tables for Al4S capacity and global scientific inequality, 2015-2025

This dataset package contains deduplicated OpenAlex-derived parquet tables constructed for bibliometric and science-of-science analyses of Al4S capacity and global scientific inequality research. The data are split into two complementary time-range directories: derived_tables_dedup/: the main deduplicated analytical package for publications from 2019-2025 derived_tables_dedup_pre2019/: the earlier deduplicated package for publications from 2015-2018 Together, these two directories provide a deduplicated publication-level and relational data resource covering 2015-2025. Core structure In each directory, the main entry table is works.parquet, which contains one deduplicated record per OpenAlex work_id. Most downstream tables can be linked back to this table through work_id. What the tables contain works.parquet: publication-level records, including work identifiers, DOI, title, abstract, publication year/date, document type, citation counts, domain labels, and topic metadata. authorships.parquet: author-by-work records, including author order, author identifiers, corresponding-author flags, and raw affiliation strings. work_institutions.parquet: institution-level affiliations linked to works and authors, including institution country and Global South indicators. authorship_countries.parquet: expanded author-country assignments used for country-level attribution analyses. first_author_country_credit.parquet: fractional country credit assigned to first authors. last_author_country_credit.parquet: fractional country credit assigned to last authors. authorship_country_diagnostics.parquet: diagnostics for first-author country attribution quality and missingness. last_author_country_diagnostics.parquet: diagnostics for last-author country attribution quality and missingness. work_topics.parquet: OpenAlex topic, subfield, field, and domain assignments for each work. work_concepts.parquet: OpenAlex concept annotations and scores. work_references.parquet: citation links between source works and referenced works. af_label_evidence.parquet: evidence table documenting why a work was labeled as AlphaFold-related, including title, abstract, reference, and keyword-based signals. Time-range note The split between derived_tables_dedup_pre2019/ and derived_tables_dedup/ is intentional. The former covers the earlier period 2015-2018, while the latter covers the main analytical period 2019-2025. This separation makes it easier to distinguish the pre-main-period literature from the main study corpus while preserving a consistent relational structure across both periods. How this dataset is intended to be used This package is designed for reproducible analyses of AlphaFold-related publication output, collaboration, country participation, institutional affiliation, topical composition, and citation structure. In most workflows, works.parquet should be used as the starting table, with the other parquet files serving as relational extensions. Note on AlphaFold labeling AlphaFold-related works are identified through a multi-signal evidence strategy. A work can be marked as AlphaFold-related based on evidence from titles, abstracts, references to core seed papers, or keyword/resource patterns. The table af_label_evidence.parquet preserves these labeling signals for auditability and reuse.