Citation Network Analysis
Building and analysing networks of authors and papers from open bibliographic metadata, to describe how a field’s literature is structured.
What it is
The pipeline has four stages.
Metadata retrieval. OpenAlex supplies open bibliographic records — authors, affiliations, references, concepts. The group uses no private or licensed datasets. Because OpenAlex updates continuously, the documented rule is to cache the metadata pull per analysis run; without caching, the same script produces different results on different days and nothing is reproducible.
Author disambiguation. The hard part. The same person appears under name variants and different people share names — acute for common surnames and for transliterated names, which matters directly for Turkish-language authorship. Disambiguation is never perfect, and the documented rule is to report the ambiguity rate alongside any author-level metric rather than presenting counts as exact.
Network construction. Two distinct graphs answer different questions. A co-authorship network links authors who published together, describing collaboration structure. A citation network links papers or authors by citation, describing influence flow. Citation networks are directed and largely acyclic in time; co-authorship networks are undirected. Conflating them produces meaningless metrics.
Analysis and visualisation. Community detection, centrality measures, and temporal trends, plotted with igraph and ggraph in R and D3 on the published site. A caution worth carrying: centrality measures are sensitive to how the network was bounded — restrict the corpus differently and the “central” authors change.
Why it matters for my work
Bibliometrics — Who Cites Who is an active project with a published companion site (whociteswho.patoloji.dev).
A correction, from reading the code on 2026-07-26. This page previously said the WHO-CNS strand “applies the same machinery” as the OpenAlex-based work above. It does not, and the difference is the interesting part. WHO CNS — Who Cites Who does not run a literature search at all: its corpus is the printed bibliographies of four WHO CNS classification editions, extracted by OCR and resolved reference by reference — 10,950 citation instances across 6,654 unique references. OpenAlex, PubMed and Lens.org are used to enrich references that already exist in the corpus, never to define it.
That inverts the usual bounding problem. There is no decision about how wide to cast the net, because the net is the book’s own reference list; what replaces it is a resolution problem, since not every printed reference can be matched to an indexed record. Asking who writes the classification the field then uses remains a genuinely interesting question, and it needs no patient data at all.
How it connects
Bibliometrics — Who Cites Who — the project page with the repo inventory and site links.
OpenAlex Pathology Metadata — the dataset page for the underlying source and its caveats.
Report Text Extraction — a sibling text-mining pipeline; both hinge on entity resolution, one for authors and one for clinical fields.
WHO CNS — Who Cites Who — the worked example of the reference-extraction variant, and the source of the denominator discipline below.
Coverage Bias — the statistical problem that resolution rates create, and the three defences WHO CNS — Who Cites Who applies against it.
Open questions
- What is the current author-disambiguation ambiguity rate on the group’s corpus? Not recorded. [unverified]
How is the corpus bounded for the WHO-CNS strand?Answered. Not by journal, concept or citation distance — the corpus is the four editions’ own printed bibliographies. The binding constraint is reference resolution, not corpus bounding, and resolution rate varies by edition and publication era. See Coverage Bias.Is the cached metadata snapshot archived with each analysis?Answered for WHO CNS — Who Cites Who. Yes: the Lens.org pull is a persistent public collection snapshotted to file,DATA_PROVENANCE.mddeclares which tables are authoritative versus stale, and the curated citation table has a frozen golden copy that the build auto-restores. Still unrecorded for the OpenAlex-based strand. [unverified]- Does the group’s other bibliometric work declare denominators and coverage the way WHO CNS — Who Cites Who does? If not, that is the cheapest available upgrade — see Coverage Bias.