Jacob A. Dalsgaard, Filipi N Silva, Jin Ai
9 min
Abstract
Funding acknowledgments in scholarly publications provide large-scale trace data on organizations that support scientific research. We present a dataset for linking global science funding organizations to research publications by systematically disambiguating unique funding acknowledgment strings extracted from publication metadata. Funder names are matched to standardized organizational identifiers using a multi-stage pipeline that combines lexical normalization, similarity-based clustering, rule-based matching, named entity recognition assistance, and manual validation. The resulting dataset links 1.9 million unique funder strings to canonical organization identifiers and records match types and unresolved cases to support transparency. Technical validation includes paper-level comparisons across bibliometric sources and manual verification against full-text acknowledgment sections, with reported recall and precision metrics. This dataset supports analyses of funding flows, institutional funding portfolios, regional representation, and concentration patterns in the global research system.
Alex: Right—and that matters for seeing funding gaps in places like Africa or Latin America?
Sam: Precisely. One notable pattern is how just 1311 funders back about 80% of all publications—mostly big players from wealthy countries. This concentration highlights why better matching reveals imbalances that fragmented data hides, like underfunding for regional work. The approach keeps ambiguities flagged for users, ensuring trust in the big-picture maps it enables.
Alex: So the real value is finally quantifying those flows without the old blind spots...
Sam: Yes. It provides a solid base for studying how funding steers science—priorities, regions, even shifts from governments to philanthropies—without overclaiming perfection.
Alex: To make those maps work, they needed a reliable directory of funders first, right? Like building a master list before doing the matching.
Sam: Precisely. They started by combining data from OpenAlex, which tracks institutions and funders with details like names and locations, and the Research Organization Registry, or ROR—a stable directory with unique codes for research groups worldwide. To merge them cleanly, the team first cleaned up names by switching to all lowercase letters and pulling out key web addresses from homepages as steady identifiers. They linked records step by step: first matching exact unique codes from ROR, then from Wikidata, and finally by shared web domains plus country info.
Alex: That stepwise linking sounds methodical—like double-checking addresses before filing mail. But what if duplicates still slipped through?
Sam: They scanned for leftovers by looking for records with the same cleaned name and country. When found, they picked one main version, favoring those from OpenAlex's funder list or ones with fuller details like grant numbers. All name variations, including acronyms, got bundled into extra fields for later searches. This creates a single, rich reference list that handles real-world messiness, like translations or abbreviations.
Alex: With that directory ready, how do they group the messy names from papers without false connections?
Sam: First, they simplify every funder name from papers to lowercase and spot acronyms, say pulling "NSF" from "National Science Foundation (NSF)." Then, to group similar ones—like variants from typos or formats—they use a trick that spots near-matches fast: it breaks names into short character chunks and compares how many overlap, linking those above 95% alike into clusters. Think of it like sorting similar-sounding player names on a sports roster.
Alex: Efficient grouping first, then verify. And for clusters without a clear directory hit?
Sam: They run ordered checks: exact name match first, then alternate names, partial overlaps covering at least half the text—like prefix, suffix, or middle substring—and finally acronyms. A solid match gets the directory's unique ID spread to the whole cluster. Tough cases get human checks for common ones, AI name-spotting for medium ones, or a final close-similarity scan—keeping options open if ambiguous, using clues like author countries.
Alex: So the whole system stays transparent by flagging uncertainties... That logic really tackles the fragmentation head-on.
Sam: They tested it two main ways. First, they compared funding lists across databases for the same papers—pulling millions indexed in Web of Science alongside OpenAlex or Dimensions, then seeing how many funders overlapped using shared IDs. Their cleaned Web of Science data captures all funders from the others in about two-thirds of cases, a clear edge.
Alex: That's automated—any hands-on proof?
Sam: Yes, they manually reviewed 250 random papers: grabbed the full text, read the thank-you sections word-for-word, and tallied matches against their dataset. It caught 78% of true funders with just 4% false ones—better than OpenAlex or Dimensions in the same test. They share all notes online for anyone to verify.
Alex: And for rarer funders, did frequency affect accuracy?
Sam: Exactly as you'd expect: common names matched near-perfectly, while one-offs had higher unmatched rates around 70-75%, left open intentionally. This coverage lets users study big flows reliably, like how databases vary most on Global South groups.
Alex: With that evidential base, what do the comparisons across databases reveal about coverage gaps—like by country?
Sam: They looked at maps of funding coverage since 2010, first by author countries from paper affiliations. Web of Science and Dimensions show higher shares in North America, China, and Europe, while OpenAlex has lower overall, and Dimensions dips in many Global South spots. Switching to funder locations, Web of Science spreads wider, catching more in Africa, South America, Southeast Asia; Dimensions sticks to high-income areas.
Alex: So the choice of database skews what you see geographically. Like, WoS picks up regional funders others miss?
Sam: Yes. These patterns flag biases users must weigh—for instance, Dimensions might underplay Global South flows because it captures fewer local funders. The paper documents this as technical validation, not judgment, to guide analysis choices. They ranked funders by publication count: just 1311 handle 80% of them, so top ones dominate, while many small ones cover niche areas, early careers, or regional work.
Alex: Small funders punch above in diversity, even if low volume... Makes sense for balanced views.
Sam: Precisely. For ambiguities, they add a paper-level step: pick best candidate using author-country match first, then funder prevalence in similar papers. It resolves per instance transparently, replicable via their software. But the method notes real limits: about 28% of strings stay unmatched, mostly rare ones or Global South funders.
Alex: Gaps persist where data's thinnest, which is telling in itself. Still, they share everything openly?
Sam: Exactly. The dataset's on Zenodo as a simple file with strings, matched IDs from OpenAlex and ROR, frequencies, and process notes. Code's on GitHub for anyone to run or tweak, stressing re-checks for new data.
Alex: Users get tools to build on, aware of the edges... Ties back to those concentrations.
Sam: Precisely. This first clear mapping reveals how funding concentrates yet diversifies science amid volatile budgets, rising collaborations, and sector shifts. It equips studies on flows by region or type, a meaningful step for policy without overclaiming completeness.
Alex: Yeah, grounds the big questions on who shapes discovery... Thanks, Sam—that's a solid look at linking funding to science.
Sam: My pleasure, Alex. Thanks for listening to ResearchPod.