WP44: bibliography merge script (599 entries, 15 duplicate keys, 0 DOI conflicts)

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Simon-Pierre Boucher committed 4 h ago (Nov 3, 2026) parent 96bb3ab Browse files

5 changed files +566 −2

modified .gitignore +2 −0

		`@@ -52,3 +52,5 @@ manuscript/*.bbl`
52	52	manuscript/*.blg
53	53	manuscript/*.synctex.gz
54	54	.venv/
	55	+references_all.bib
	56	+manuscript/references_merge_report.txt

modified manuscript/main.tex +1 −1

		`@@ -82,7 +82,7 @@`
82	82	\input{sections/part11_discussion}
83	83
84	84	\clearpage
85		−\bibliography{references}
	85	+\bibliography{references_all}
86	86
87	87	\begin{appendices}
88	88	\input{appendices/appA_taxonomy}

added manuscript/merge_bibs.py +31 −0

		`@@ -0,0 +1,31 @@`
	1	+#!/usr/bin/env python3
	2	+"""Merge manuscript/references.bib (main, Agent B-lit) with the writers' references_part*.bib into references_all.bib.
	3	+First occurrence wins (main bib first); duplicate keys are reported in references_merge_report.txt with a flag when the
	4	+DOI differs (a real conflict to resolve by hand). main.tex uses \\bibliography{references_all}."""
	5	+import glob, re, os
	6	+H = os.path.dirname(os.path.abspath(__file__))
	7	+files = [os.path.join(H, "references.bib")] + sorted(glob.glob(os.path.join(H, "references_part*.bib")))
	8	+seen = {}; out = []; dups = []
	9	+for f in files:
	10	+ txt = open(f, encoding="utf-8").read()
	11	+ for m in re.finditer(r"@(\w+)\s\{\s([^,\s]+)\s*,", txt):
	12	+ start = m.start(); depth = 0; i = m.end() - 1
	13	+ # find matching closing brace of the entry
	14	+ j = txt.find("{", start); depth = 0; k = j
	15	+ while k < len(txt):
	16	+ if txt[k] == "{": depth += 1
	17	+ elif txt[k] == "}":
	18	+ depth -= 1
	19	+ if depth == 0: break
	20	+ k += 1
	21	+ entry = txt[start:k + 1]; key = m.group(2)
	22	+ doi = re.search(r"doi\s=\s[{\"]([^}\"]+)", entry, re.I)
	23	+ doi = doi.group(1).strip().lower() if doi else None
	24	+ if key in seen:
	25	+ dups.append((key, os.path.basename(f), "DOI_CONFLICT" if (doi and seen[key] and doi != seen[key]) else "same"))
	26	+ continue
	27	+ seen[key] = doi; out.append(entry)
	28	+open(os.path.join(H, "references_all.bib"), "w", encoding="utf-8").write("% merged by merge_bibs.py — do not edit; edit the source bib files\n\n" + "\n\n".join(out) + "\n")
	29	+with open(os.path.join(H, "references_merge_report.txt"), "w") as fh:
	30	+ fh.write(f"entries: {len(out)} from {len(files)} files\n"); [fh.write(f"DUP {k} in {f}: {st}\n") for k, f, st in dups]
	31	+print(len(out), "entries;", len(dups), "duplicates;", sum(1 for d in dups if d[2] == "DOI_CONFLICT"), "DOI conflicts")

modified manuscript/sections/part03_data.tex +316 −1

		`@@ -1 +1,316 @@`
1		−% part03_data — to be written (Phase 7)
	1	+% ============================================================================
	2	+% Part III — Data and Systematic Review Protocol
	3	+% Written by Agent L (Part III instance), 2026-11-03. Every corpus number is a macro from tables/numbers.tex
	4	+% (generated by tables/make_numbers.py from data/metadata/prisma_counts.json and linkage_report.json).
	5	+% Sources of truth: RESEARCH_PLAN.md §2–§3, protocol/PROTOCOL.md, protocol/amendments.csv,
	6	+% reports/PHASE0_FEASIBILITY.md, data/raw/README.md, ingestion/*.py, systematic_review/define_corpus.py.
	7	+% ============================================================================
	8	+
	9	+\section{Information sources and conditions of access}
	10	+\label{sec:data-sources}
	11	+
	12	+A field-level mapping study stands or falls with the bibliographic databases on which it is built. The choice
	13	+of sources determines which works can be observed at all, how reliably they can be dated and typed, whether
	14	+citation links are available, and under what licence derived data may be redistributed. This part documents
	15	+these choices for WP44 and reports, stage by stage, how the mapping corpus was assembled from the raw
	16	+sources. The exposition follows the order of the pipeline (identification, linkage, deduplication,
	17	+eligibility, scope, dating) and closes with the quality indicators and the research-integrity procedures that
	18	+govern every later analysis. The reporting structure is modelled on the PRISMA 2020 items that apply to a
	19	+mapping study \citep{page2021prisma,page2021prismaee}; items concerned with risk of bias and effect measures
	20	+have no counterpart in a bibliometric design and are replaced, as PRISMA-ScR does for scoping reviews
	21	+\citep{tricco2018prismascr}, by indicators of metadata quality and coverage (Section~\ref{sec:data-quality}).
	22	+The search-reporting recommendations of PRISMA-S \citep{rethlefsen2021prismas} are followed to the extent
	23	+that a database-scope strategy without keyword queries permits.
	24	+
	25	+\subsection{Why a multi-source design}
	26	+
	27	+No single database indexes computer science completely, and the gaps are not random. The venue-centred
	28	+coverage of DBLP, the publisher-centred coverage of Crossref, the preprint-centred coverage of arXiv and the
	29	+aggregating coverage of OpenAlex each miss a distinct part of the literature and type the part they do cover
	30	+in distinct ways. Comparative studies of bibliographic data sources have documented large differences in
	31	+coverage between Web of Science, Scopus, Dimensions, Crossref and Microsoft Academic
	32	+\citep{visser2021comparison,martinmartin2021comparison}, and the journal-level coverage of the two commercial
	33	+indexes is known to under-represent conference proceedings, which carry a large share of computer-science
	34	+output \citep{mongeon2016coverage}. For the open successor of Microsoft Academic, OpenAlex, the first
	35	+independent assessments report strong coverage of works with DOIs together with weaker and uneven metadata on
	36	+affiliations, document types and references \citep{alperin2024openalex,culbert2025openalex,
	37	+scheidsteger2023openalex,delgadoquiros2024completeness}. These findings motivate the design adopted here: one
	38	+source supplies the backbone (OpenAlex), a second supplies the definition of what counts as computer science
	39	+(DBLP), a third supplies the preprint layer and its dates (arXiv), and a fourth checks the coverage of a
	40	+subfield with a dense, curated venue structure (the ACL Anthology). Crossref is used for verification only.
	41	+
	42	+\subsection{OpenAlex}
	43	+
	44	+OpenAlex \citep{priem2022openalex} is the backbone of the corpus. It supplies, for every work, the
	45	+identifiers (OpenAlex~W identifier, DOI, arXiv identifier where present), the publication date and type, the
	46	+title and an inverted-index abstract where available, the venue (source) and its type, the authorships with
	47	+institutions and countries, the topic assignment at three levels (field, subfield, topic) with a primary
	48	+topic, the list of referenced works (the citation graph), the open-access status and licence, and the
	49	+field-weighted citation impact together with yearly citation counts. OpenAlex is released under the CC0
	50	+public-domain dedication, which permits redistribution of record-level derived data.
	51	+
	52	+Two access routes exist and both were used, for different purposes. The public snapshot is a versioned bulk
	53	+release in gzipped JSON Lines, published in a public S3 bucket; the release used here is dated
	54	+2026-10-14 and comprises 1{,}898 part files for the \emph{works} entity, 675.3~GB compressed, with a manifest
	55	+listing the record count of each file. The snapshot was streamed from the primary compute node, each file's
	56	+content length checked against the manifest and a SHA-256 of the compressed bytes recorded in a SQLite ledger
	57	+before parsing; every one of the 1{,}898 files parsed without error and the sum of the per-file record counts
	58	+equals the manifest total of \LinkSlimTotal{} works. Two outputs were written during this single pass. The
	59	+first is the full JSON record of every work carrying \emph{any} topic in OpenAlex field~17 (Computer Science)
	60	+with a publication year between 2018 and 2026, stored as Zstandard-compressed JSON Lines and partitioned by
	61	+the snapshot's own date partitions (\LinkOaCsRecords{} records, about 20~GB). The second is a \emph{slim}
	62	+table of every work in the snapshot (identifier, DOI, year, type, primary topic, subfield, field and domain,
	63	+citation count and a few flags), kept in Parquet so that works cited by computer-science papers but published
	64	+in other fields can be resolved for the interdisciplinary analyses of Part~VIII (decision D2 of
	65	+\texttt{RESEARCH\_PLAN.md}). Records flagged by OpenAlex as XPAC expansion records are retained at this stage
	66	+with their flag so that the exclusion decision can be taken, and counted, later (Section~\ref{sec:data-eligibility}).
	67	+The smaller entity snapshots (topics, subfields, fields, domains, sources, institutions, publishers, funders,
	68	+concepts) were downloaded the same day with a per-file SHA-256 ledger; author entities were not downloaded
	69	+because the authorship data embedded in each work record suffice for the analyses planned.
	70	+
	71	+The second route is the REST API. Without an API key the public endpoint operates under a shared daily budget
	72	+of USD~0.10, list-and-filter calls are priced at USD~0.0001 each, and cursor paging is unavailable, so that a
	73	+single filtered query cannot return more than 10{,}000 results through basic paging. Single-record lookups by
	74	+identifier are free. The API was therefore used for two narrowly defined tasks. First, an incremental harvest
	75	+of works with any computer-science topic and a publication date between 2026-10-01 and 2026-11-02, intended
	76	+to capture records added after the snapshot was cut; the harvest is sliced by publication day and, where a day
	77	+exceeds the paging limit, split into records with a computer-science primary field and records without one.
	78	+Raw JSON pages are stored verbatim under a dated directory with a ledger of counts and cost, and the job
	79	+resumes on the following day when the budget is exhausted. Second, free single-record lookups by DOI serve as
	80	+one of the verification routes for the bibliography of this manuscript (Section~\ref{sec:data-integrity}).
	81	+A free API key, which would raise the daily budget, requires the principal investigator to act on an e-mail
	82	+and was deliberately not obtained without instruction; the mandate forbids assuming credentials that have not
	83	+been tested.
	84	+
	85	+\subsection{DBLP}
	86	+
	87	+The DBLP computer science bibliography \citep{ley2002dblp,ley2009dblp} is a curated, venue-centred index that
	88	+has for three decades defined, in practice, the set of journals and conference series regarded as computer
	89	+science. It is used here in that capacity: a work indexed by DBLP is in computer science by construction, and
	90	+DBLP venue streams feed the taxonomy of Part~IV. The complete XML dump (\texttt{dblp.xml.gz}, dated 2026-11-02
	91	+22:39~UTC, 1.12~GB) was downloaded together with its DTD and MD5 checksum, the checksum verified, and the file
	92	+parsed with a validating XML parser into one row per publication record (articles, conference papers,
	93	+proceedings volumes, books, book chapters, doctoral and master's theses, data records; the \texttt{www}
	94	+person records are skipped). The whole archive was kept so that per-year counts are available for the
	95	+historical baseline and for context. The DBLP dump is released under the Open Data Commons Attribution licence
	96	+(ODC-BY~1.0). DBLP contributes, in addition to the scope definition, author person identifiers, venue streams,
	97	+and DOI links for most records; the \texttt{ee} fields were also scanned for arXiv URLs, with a result reported
	98	+in Section~\ref{sec:data-linkage}.
	99	+
	100	+\subsection{arXiv}
	101	+
	102	+The arXiv e-print repository \citep{ginsparg2011arxiv} is the main preprint channel of computer science and,
	103	+for the period under study, the place where a large share of the most cited work first appeared. Its
	104	+metadata are harvested through the OAI-PMH interface (\texttt{set=cs}, \texttt{metadataPrefix=arXiv}) in
	105	+monthly datestamp windows from January 2018 to 2 November 2026. The harvester waits at least 3.5~s between
	106	+requests and honours \texttt{Retry-After} headers; every raw XML page is kept verbatim as the immutable
	107	+snapshot and parsed afterwards into a single Parquet table. Two properties of OAI-PMH matter for what follows.
	108	+The \texttt{from}/\texttt{until} parameters filter on the record's \emph{datestamp}, the time of its last
	109	+update, so that a monthly window contains every record updated in that month, not every record first
	110	+submitted then; duplicates across windows are resolved at parse time with the latest datestamp winning. The
	111	+\texttt{created} element gives the first submission date, which is the date used throughout this study for
	112	+preprints. The harvest returned \LinkArxivCsRecords{} distinct records, of which
	113	+\LinkArxivCreatedTwoZeroOneEightTwoZeroTwoSix{} were first submitted in 2018--2026; the remainder are older
	114	+e-prints updated during the window. The arXiv metadata are released for reuse under the repository's metadata
	115	+terms; e-prints themselves are governed by per-paper licences and are downloaded, for the high-priority set
	116	+only, within those licences and at the repository's rate limits. The use of arXiv as a research dataset, and
	117	+the caveats attached to its category labels and versioning, are discussed by \citet{clement2019arxiv}.
	118	+
	119	+\subsection{ACL Anthology}
	120	+
	121	+The ACL Anthology is the curated archive of the computational-linguistics community and has itself served as a
	122	+bibliometric corpus \citep{radev2013acl}. Its BibTeX export with abstracts (\texttt{anthology+abstracts.bib.gz},
	123	+46.7~MB, SHA-256 recorded) was downloaded on 2026-11-02 and parsed into a table of 138{,}796 entries, of which
	124	+\LinkAclRecords{} are conference papers or journal articles and \LinkAclTwoZeroOneEightTwoZeroTwoSix{} fall in
	125	+2018--2026. The Anthology plays a diagnostic role: because its venue structure is complete and hand-maintained,
	126	+the share of its entries that the backbone can be matched to is a direct measure of OpenAlex coverage for a
	127	+subfield that publishes overwhelmingly in conference proceedings. It also supplies abstracts for those entries
	128	+in which OpenAlex has none. The export is licensed CC~BY~4.0.
	129	+
	130	+\subsection{Crossref, Semantic Scholar and other sources}
	131	+
	132	+Crossref \citep{hendricks2020crossref} is the registration agency behind most DOIs in the corpus and offers
	133	+open metadata through a REST API with a polite pool. It is not used to build the corpus, whose backbone
	134	+already carries Crossref-derived metadata through OpenAlex, but for three verification purposes: resolution of
	135	+every DOI cited in this manuscript, a cross-check of publisher-assigned document types on samples, and a
	136	+cross-check of citation counts (\texttt{is-referenced-by-count}) on a sample of highly cited works in
	137	+Part~IX.
	138	+
	139	+The Semantic Scholar Academic Graph \citep{kinney2023s2} was listed in the mandate among the priority sources
	140	+because of two signals it alone provides, the influential-citation classification and citation intent. At the
	141	+time of the feasibility audit, anonymous requests to its API returned HTTP~429 with an instruction to apply for
	142	+a key, and the bulk-dataset endpoint requires a key outright. In keeping with the rule that inaccessible data
	143	+be recorded rather than circumvented, Semantic Scholar is recorded as unavailable (decision D3) and no analysis
	144	+in this manuscript uses influential-citation or citation-intent signals; where Part~VIII discusses the
	145	+strength of citation links, it does so with OpenAlex reference lists and the knowledge-graph extraction of
	146	+Part~IV, whose limits are stated there. The Microsoft Academic Graph \citep{wang2020mag}, from which OpenAlex
	147	+inherited much of its historical coverage, was discontinued at the end of 2021 and is not an independent source
	148	+for the period studied. Kaggle credentials for the arXiv snapshot distributed there were absent and were not
	149	+needed, OAI-PMH being sufficient. Benchmark repositories and code-hosting platforms enter only in the
	150	+reproducibility and benchmark analyses scoped in Part~IV and are not part of the bibliographic corpus.
	151	+
	152	+Table~\ref{tab:data-sources} summarises the sources, the access route, the licence and the role of each.
	153	+
	154	+\begin{table}[htbp]
	155	+\centering\small
	156	+\caption{Information sources of the WP44 corpus (all accessed 2026-11-02/03).}
	157	+\label{tab:data-sources}
	158	+\begin{tabularx}{\textwidth}{lXll}
	159	+\toprule
	160	+Source & Access and snapshot & Licence & Role \\
	161	+\midrule
	162	+OpenAlex works snapshot & public S3 bucket, release 2026-10-14, 1{,}898 JSONL files, \LinkSlimTotal{} works & CC0 & backbone: identifiers, dates, types, topics, authorships, references, abstracts, OA status \\
	163	+OpenAlex REST API & no key; USD~0.10 per day shared budget; increment 2026-10-01 to 2026-11-02; single lookups & CC0 & post-snapshot increment; reference verification \\
	164	+DBLP & XML dump 2026-11-02 22:39~UTC, MD5 verified & ODC-BY~1.0 & computer-science scope; venue streams; author identifiers \\
	165	+arXiv & OAI-PMH set \texttt{cs}, 2018-01 to 2026-11, raw pages kept & arXiv metadata terms & preprint layer; first-submission dates; categories \\
	166	+ACL Anthology & BibTeX export with abstracts, 2026-11-02 & CC~BY~4.0 & coverage check for NLP venues; abstracts \\
	167	+Crossref & REST API, polite pool & open metadata & verification of DOIs, types and citation counts \\
	168	+Semantic Scholar & API key required; anonymous calls refused & --- & \emph{unavailable, recorded} \\
	169	+\bottomrule
	170	+\end{tabularx}
	171	+\end{table}
	172	+
	173	+\section{Identification: what each source contributes}
	174	+\label{sec:data-identification}
	175	+
	176	+There is no keyword search in this study. A mapping of an entire discipline cannot be defined by a query
	177	+string without importing the vocabulary of whoever wrote it; it is defined instead by database scope filters
	178	+that are themselves documented and reproducible, which is the strategy recommended for mapping studies in
	179	+software engineering and adopted here for computer science as a whole
	180	+\citep{petersen2008mapping,petersen2015mapping}. The identification stage (S0 in the notation of the research
	181	+plan) therefore consists of four scope filters, one per source, applied to immutable snapshots whose dates and
	182	+hashes are recorded in \texttt{data/raw/README.md}.
	183	+
	184	+\subsection{Scope filters}
	185	+
	186	+The OpenAlex filter retains every work with at least one topic, primary or secondary, in field~17 and a
	187	+publication year in $[2018, 2026]$. The choice of \emph{any} topic rather than the primary topic is deliberate
	188	+and conservative at this stage: a work whose primary topic lies in, say, statistics or electrical engineering
	189	+but which carries a secondary computer-science topic is a candidate, and the decision whether it belongs to
	190	+the core or to the penumbra is deferred to the eligibility stage, where it is made explicit and counted
	191	+(Section~\ref{sec:data-core}). The filter yields
	192	+\PrismaIdentifiedOpenalexCsTopicTwoZeroOneEightTwoZeroTwoSixPassOne{} records in the first pass over the
	193	+snapshot. The DBLP filter retains the publication records of type article, inproceedings, incollection, book,
	194	+phdthesis and data with a year in $[2018, 2026]$: \PrismaIdentifiedDblpTwoZeroOneEightTwoZeroTwoSix{} records.
	195	+The arXiv filter retains the records of set \texttt{cs} whose first submission falls in the window:
	196	+\PrismaIdentifiedArxivCsTwoZeroOneEightTwoZeroTwoSix{} records. The ACL filter retains the conference papers
	197	+and journal articles of the Anthology with a year in the window:
	198	+\PrismaIdentifiedAclTwoZeroOneEightTwoZeroTwoSix{} records. The four counts overlap heavily, since most DBLP
	199	+and ACL records and a majority of arXiv records also exist in OpenAlex; the overlap is resolved by record
	200	+linkage, not by summation, and no total of identified records is reported as if the sources were disjoint.
	201	+
	202	+\subsection{The OpenAlex topic classifier as a scope instrument}
	203	+
	204	+Because the OpenAlex filter relies on an automated topic classifier, its properties bear on everything
	205	+downstream. OpenAlex assigns to each work up to three topics from a fixed list of 4{,}516, organised in 252
	206	+subfields, 26 fields and four domains; the assignment is made by a model applied to the title, abstract and
	207	+venue, and the field label of a work is inherited from its topics rather than assigned directly. Two
	208	+consequences follow. A work can be labelled computer science because one of its topics belongs to field~17
	209	+even when its content is only marginally computational, and conversely a work in a computer-science venue can
	210	+receive no computer-science topic when its abstract is missing or its vocabulary is domain-specific. The first
	211	+error inflates the candidate set with what will be called the penumbra; the second removes genuine
	212	+computer-science works from the first-pass candidate set. DBLP is used to correct the second error, which is
	213	+why the linkage stage has two passes (Section~\ref{sec:data-linkage}). The literature on OpenAlex has so far
	214	+examined its coverage and metadata completeness rather than the accuracy of its topic labels
	215	+\citep{alperin2024openalex,delgadoquiros2024completeness}; the present study treats the labels as a scope
	216	+instrument whose errors are bounded by the DBLP and arXiv memberships and quantified in
	217	+Section~\ref{sec:data-quality}, not as ground truth.
	218	+
	219	+\subsection{Verification of the scale against the mandate}
	220	+
	221	+The mandate set planning targets of more than one million candidate records and more than one hundred
	222	+thousand deeply classified records, subject to availability. The feasibility audit established, before any
	223	+ingestion, that the OpenAlex API alone reported more than six million works with a computer-science primary
	224	+field in 2018--2026 and that the DBLP dump held more than four million records in the same years; the targets
	225	+were therefore attainable without recourse to any restricted source. The audit also established what was
	226	+\emph{not} attainable: a Semantic Scholar key, an OpenAlex key, Kaggle credentials and the AWS command-line
	227	+tool were all absent, and the ingestion was designed around their absence (HTTPS streaming of the public
	228	+bucket, basic paging of the public API). The record of this audit is \texttt{reports/PHASE0\_FEASIBILITY.md}.
	229	+
	230	+\section{Record linkage across sources}
	231	+\label{sec:data-linkage}
	232	+
	233	+The four candidate sets describe overlapping populations with different identifiers. Record linkage (stage S1)
	234	+establishes, for each DBLP, arXiv and ACL record, the OpenAlex work it describes, so that one node per work
	235	+can be formed with the union of the metadata. The methods are deterministic and ordered by reliability; the
	236	+probabilistic framework of \citet{fellegi1969linkage} is not needed because the available keys (DOI, arXiv
	237	+identifier) are close to exact, and the residual title-based step is restricted to blocks where it is
	238	+unambiguous. The implementation follows the standard decomposition of data matching into key normalisation,
	239	+blocking, comparison and classification \citep{christen2012matching}; Part~IV gives the formal statement of the
	240	+keys and of the precision-estimation design, and this section reports what was done and what it produced.
	241	+
	242	+\subsection{Keys and matching order}
	243	+
	244	+All keys are normalised before comparison. DOIs are lower-cased. arXiv identifiers are lower-cased and
	245	+stripped of their version suffix (\texttt{v2}, \texttt{v3}, \ldots), so that all versions of an e-print map to
	246	+one identifier. Titles are reduced to a normalised string by lower-casing and deleting every character that is
	247	+not a letter or digit; the normalised title is used only when it has at least 25 characters, is unique within
	248	+its source, and is unique within OpenAlex, which is the blocking rule that keeps the title step from
	249	+producing spurious links between generic titles.
	250	+
	251	+For DBLP, the order is: (i) DOI equality, (ii) arXiv identifier equality, restricted to records not yet linked
	252	+on either side, (iii) normalised title equality with a year difference of at most one, restricted to records
	253	+not yet linked on either side. The DOI step linked \LinkLinkDblpOaDoi{} records and the title step
	254	+\LinkLinkDblpOaTitleYear{}; the arXiv step linked none, because the extraction of arXiv URLs from the DBLP
	255	+\texttt{ee} fields returned no candidate (\LinkDblpWithArxiv{} records with an extracted identifier). The most
	256	+plausible reason is that DBLP now records the arXiv version of its CoRR entries as a DataCite DOI of the form
	257	+\texttt{10.48550/arXiv.*}, which the DOI step already handles; the step is kept in the pipeline because other
	258	+dumps may differ. In total \LinkDblpLinkedDistinct{} of the \LinkDblpTwoZeroOneEightTwoZeroTwoSix{} DBLP
	259	+records in the window were linked to an OpenAlex work after the first pass, leaving \LinkDblpUnlinked{}
	260	+unlinked. The composition of the unlinked set is informative: \LinkDblpUnlinkedByTypeArticle{} journal
	261	+articles, \LinkDblpUnlinkedByTypeInproceedings{} conference papers, \LinkDblpUnlinkedByTypePhdthesis{}
	262	+doctoral theses, \LinkDblpUnlinkedByTypeData{} data records, \LinkDblpUnlinkedByTypeIncollection{} chapters
	263	+and \LinkDblpUnlinkedByTypeBook{} books. Of these, \LinkDblpUnlinkedWithDoiInSlim{} carry a DOI that
	264	+\emph{does} exist in the slim table of all OpenAlex works, which means the work is in OpenAlex but was not
	265	+caught by the computer-science topic filter; this is the second error of the topic classifier described above
	266	+and the reason for the second pass.
	267	+
	268	+For arXiv, the order is: (i) arXiv identifier equality against the identifier OpenAlex records for the work,
	269	+(ii) DOI equality for e-prints that carry a journal DOI, (iii) normalised title equality with a year
	270	+difference of at most two, the wider tolerance reflecting the typical lag between a preprint and its published
	271	+version \citep{lin2020preprints}. The steps linked \LinkLinkArxivOaArxivId{}, \LinkLinkArxivOaDoi{} and
	272	+\LinkLinkArxivOaTitleYear{} records respectively, for \LinkArxivLinkedDistinct{} distinct e-prints linked.
	273	+For the ACL Anthology, the order is: (i) DOI equality, (ii) equality between the Anthology identifier and the
	274	+identifier embedded in an \texttt{aclanthology.org} landing-page URL among the work's OpenAlex locations,
	275	+(iii) normalised title equality with a year difference of at most one. The steps linked
	276	+\LinkLinkAclOaDoi{} records by DOI and \LinkLinkAclOaTitleYear{} by title, with the URL step contributing the
	277	+remainder, for \LinkAclLinkedDistinctTwoZeroOneEightTwoZeroTwoSix{} distinct 2018--2026 entries linked out of
	278	+\LinkAclTwoZeroOneEightTwoZeroTwoSix{}; the residual is examined in Section~\ref{sec:data-quality} as a
	279	+measure of OpenAlex coverage of conference-heavy subfields.
	280	+
	281	+\subsection{The second snapshot pass}
	282	+
	283	+A DBLP record whose DOI resolves to an OpenAlex work outside the topic filter is, by the scope rule, a
	284	+computer-science work that the first pass did not load. The same situation arises for arXiv e-prints whose
	285	+journal DOI exists in OpenAlex without a computer-science topic. The linkage script therefore writes the list
	286	+of OpenAlex identifiers that are needed but not loaded: \LinkNeededOaIdsSecondPassDblpDoi{} identifiers from
	287	+DBLP DOIs and \LinkNeededOaIdsSecondPassArxivDoi{} from arXiv DOIs. A second pass over the works snapshot,
	288	+restricted to this identifier list, retrieves the full records of these works; they are normalised into the
	289	+same Parquet tables with a \texttt{\_p2} suffix, the DuckDB views are rebuilt over the union of both passes,
	290	+and linkage is re-run so that the DBLP-linked set gains the works that the topic classifier had missed. The
	291	+second pass is what makes the DBLP membership rule of Section~\ref{sec:data-core} operational: without it,
	292	+``DBLP-indexed'' would silently mean ``DBLP-indexed and labelled computer science by OpenAlex'', and the DBLP
	293	+correction would be void precisely where it matters. DBLP records with no OpenAlex counterpart after both
	294	+passes (\PrismaLinkedDblpOnlyNoOpenalexRecord{} records before the second pass) are kept in a separate table
	295	+with their year and type and are reported as such; they enter the volume counts of Part~IX only in the
	296	+sensitivity analysis that adds DBLP-only records, because they lack the abstracts, references and affiliations
	297	+on which every other indicator depends. Most of them are journal articles and conference papers with a DOI
	298	+that OpenAlex does not hold at all, together with doctoral theses, which DBLP indexes from university
	299	+repositories that OpenAlex covers unevenly.
	300	+
	301	+\subsection{Precision of the linkage}
	302	+
	303	+Exact-key links are correct whenever the key is correct in both sources; the known failure modes are DOIs
	304	+reassigned by publishers between versions, DOIs registered for a proceedings volume rather than a paper, and
	305	+e-prints whose OpenAlex record points to a different version. Title-based links can fail on near-identical
	306	+titles of distinct works (extended abstracts and full papers, workshop and conference versions). The protocol
	307	+pre-specifies the estimation of linkage precision on a sample of 500 DBLP--OpenAlex pairs stratified by method,
	308	+with automatic agreement checks (normalised-title similarity, year difference, first-author surname) and a
	309	+200-pair subsample reserved for the principal investigator (checkpoint H-1). The sampling script draws 250 DOI
	310	+links, 200 title links and 50 arXiv links with a fixed seed; since the arXiv step produced no links in this
	311	+dump, the realised sample has 450 pairs, and the precision estimate for the arXiv step is undefined rather
	312	+than reported as perfect. At the time of writing the sampling script exists but the sample has not been drawn and the
	313	+human checkpoint is unsigned; the automatic agreement rates and the human verdicts will be reported in Part~IX
	314	+and Appendix~B, and every conclusion that depends on the title-based links is labelled as pending human review
	315	+until H-1 is signed.
	316	+

modified manuscript/sections/part04_methodology.tex +216 −0

		`@@ -553,3 +553,219 @@ share of the corpus) are classified from the title and the metadata only and`
553	553	have lower recall for the keyword-dependent categories; the share of such
554	554	works is reported by year because it is not constant.
555	555
	556	+\section{Bibliometric indicators}
	557	+\label{sec:m-bibliometrics}
	558	+
	559	+The indicators of this section are the estimands of research questions RQ1,
	560	+RQ3 and RQ6. They are standard in the scientometric literature, and the
	561	+definitions follow the review of \citet{waltman2016review} where a choice had
	562	+to be made; what is specific to this study is the citation-window discipline
	563	+imposed by the eight-year span of the corpus and the explicit treatment of the
	564	+two counting conventions. The scripts are \texttt{bibliometrics/01\_volume\_series.py},
	565	+\texttt{02\_collaboration.py} and \texttt{03\_citation\_impact.py}.
	566	+
	567	+\subsection{Volume, composition and growth}
	568	+
	569	+The monthly count $N_t$ and the category counts $N_{k,t}$ of
	570	+equation~\eqref{eq:counts} are tabulated by first month, period, date precision,
	571	+type and source membership (DBLP-indexed, OpenAlex-primary, arXiv-linked), with
	572	+the numbers of works having an abstract, having at least one resolvable
	573	+reference, and being open access; the annual table adds the mean number of
	574	+authors, references and countries per work. Composition is the share
	575	+$s_{k,t}$; its annual counterpart uses the primary label on all eligible works.
	576	+Growth between two dates separated by $\Delta$ years is summarised by the
	577	+compound annual growth rate
	578	+\begin{equation}
	579	+ \mathrm{CAGR}(t_a, t_b) \;=\; \Bigl( \frac{N_{t_b}}{N_{t_a}} \Bigr)^{1/\Delta} - 1,
	580	+ \qquad \Delta = (t_b - t_a)\ \text{in years},
	581	+ \label{eq:cagr}
	582	+\end{equation}
	583	+applied to annual totals and to annual category totals; it is a descriptive
	584	+summary of two end points and says nothing about the path between them, which
	585	+is why the structural-change analysis of Section~\ref{sec:m-structural} works
	586	+on the full monthly series instead. The CAGR is computed at table-making time
	587	+from the annual counts rather than by a dedicated script; the plan lists it
	588	+among the Phase-4 indicators and it is reported for complete years only, since
	589	+2026 is right-censored (Section~\ref{sec:m-truncation}).
	590	+
	591	+\subsection{Fixed citation windows}
	592	+\label{sec:m-citations}
	593	+
	594	+OpenAlex provides, for each work, the number of citations received in each
	595	+calendar year (\texttt{counts\_by\_year}). Let $c_{w}(u)$ be the citations
	596	+received by $w$ in calendar year $u$ and $y(w)$ its publication year. The
	597	+fixed-window counts are
	598	+\begin{equation}
	599	+ c^{(12)}_w = \sum_{u \le y(w)+1} c_w(u), \qquad
	600	+ c^{(36)}_w = \sum_{u \le y(w)+2} c_w(u), \qquad
	601	+ c^{(48)}_w = \sum_{u \le y(w)+3} c_w(u),
	602	+ \label{eq:windows}
	603	+\end{equation}
	604	+so that $c^{(12)}$ counts citations in the publication year and the following
	605	+calendar year, $c^{(36)}$ in the publication year and the two following years,
	606	+and so on. The superscripts are the names used in the code and in the tables
	607	+and must be read as calendar-year windows rather than as exact month counts: a
	608	+work published in January and one published in December of the same year
	609	+receive windows that differ by eleven months in effective length. This
	610	+granularity is imposed by the data (OpenAlex does not provide citation dates
	611	+below the year) and is the same for the baseline and the post period, so it
	612	+blurs comparisons within a year but does not bias comparisons between years.
	613	+The window is the device that neutralises citation-age bias: comparing the
	614	+total citation count of a 2019 work with that of a 2025 work compares seven
	615	+years of exposure with one, whereas comparing $c^{(36)}$ compares equal
	616	+exposures. The cost is that the last window-length of the corpus cannot be
	617	+evaluated; $c^{(36)}$ is complete for publication years up to 2023 given a
	618	+snapshot taken late in 2026, and $c^{(12)}$ up to 2024. Tables flag the years
	619	+for which a window is incomplete, and the choice of window length follows the
	620	+evidence in \citet{wang2013window} that short windows rank works materially
	621	+differently from long ones in slow-moving fields, which is why two lengths are
	622	+reported rather than one.
	623	+
	624	+\subsection{Field normalisation}
	625	+
	626	+Raw citation counts differ across subfields and document types by factors that
	627	+have nothing to do with influence. The normalised citation score of a work is
	628	+its window count divided by the mean window count of its reference class
	629	+\citep{waltman2016review, waltman2019fieldnorm}:
	630	+\begin{equation}
	631	+ \mathrm{NCS}^{(h)}_w \;=\; \frac{c^{(h)}_w}{\bar c^{(h)}\bigl(y(w), \tau(w), f(w)\bigr)}, \qquad
	632	+ \bar c^{(h)}(y, \tau, f) = \frac{1}{\|\mathcal{W}_{y,\tau,f}\|}\sum_{v \in \mathcal{W}_{y,\tau,f}} c^{(h)}_v ,
	633	+ \label{eq:ncs}
	634	+\end{equation}
	635	+where $f(w)$ is the primary OpenAlex subfield of $w$ and $\mathcal{W}_{y,\tau,f}$
	636	+the set of eligible works sharing year, type and subfield. The reference class
	637	+is computed on the eligible corpus, not on all of OpenAlex, so the
	638	+normalisation is relative to computer science as delimited here: an NCS of 2
	639	+means twice the mean of comparable computer-science works, not twice the world
	640	+mean. The score is undefined when the class mean is zero (very small classes in
	641	+the most recent years) and is then left missing rather than set to zero.
	642	+Normalisation by the mean of a heavy-tailed distribution is sensitive to a few
	643	+very highly cited works in the class; the median-based alternatives discussed
	644	+by \citet{waltman2016review} are not used because the median of $c^{(36)}$ is
	645	+zero in many classes of this corpus, which would make them uninformative.
	646	+
	647	+OpenAlex's own field-weighted citation impact (FWCI) is kept as a secondary
	648	+indicator. Its documentation defines it as ``citations received divided by
	649	+citations expected for works of the same type, year, and subfield'', with a
	650	+world average of one \citep{openalex2026fwci}. It differs from
	651	+equation~\eqref{eq:ncs} in three respects: the expected count is computed over
	652	+all of OpenAlex rather than over this corpus; the citation count is the total
	653	+to date rather than a fixed window, so that FWCI values of recent works are
	654	+based on a short and unequal exposure; and the computation is OpenAlex's,
	655	+recomputed nightly, so that the value depends on the snapshot date. For these
	656	+reasons FWCI is reported for comparison and never used to select works.
	657	+
	658	+\subsection{Inequality and concentration}
	659	+
	660	+Citation inequality within a publication year is measured by the Gini
	661	+coefficient of $c^{(36)}$ among the $n$ eligible works of that year. With the
	662	+works sorted so that $c_{(1)} \le c_{(2)} \le \dots \le c_{(n)}$, the estimator
	663	+implemented in SQL is the rank form
	664	+\begin{equation}
	665	+ G \;=\; \frac{2\sum_{i=1}^{n} i\, c_{(i)}}{n \sum_{i=1}^{n} c_{(i)}} - \frac{n+1}{n},
	666	+ \label{eq:gini}
	667	+\end{equation}
	668	+which equals one minus twice the area under the Lorenz curve and lies in
	669	+$[0, 1 - 1/n]$; the coefficient is named after \citet{ceriani2012gini}'s
	670	+account of its origin. Uncited works (a large share within a three-year
	671	+window) contribute zero to the numerator and pull $G$ upward, so the share of
	672	+uncited works is reported next to it. The protocol (H5) refers to the Gini of
	673	+24-month citations; the implemented statistic uses the three-calendar-year
	674	+window $c^{(36)}$, which is complete for one more publication year than the
	675	+longer alternatives and less dominated by zeros than $c^{(12)}$. This is a
	676	+deviation from the protocol wording and is recorded as such.
	677	+
	678	+Venue concentration within a year is the Herfindahl--Hirschman index of the
	679	+distribution of works over sources,
	680	+\begin{equation}
	681	+ \mathrm{HHI}_y \;=\; \sum_{v} \sigma_{v,y}^2, \qquad
	682	+ \sigma_{v,y} = \frac{N_{v,y}}{\sum_{v'} N_{v',y}},
	683	+ \label{eq:hhi}
	684	+\end{equation}
	685	+where $N_{v,y}$ is the number of eligible works of year $y$ whose primary
	686	+source is $v$ \citep{hirschman1964paternity}; works without a source are
	687	+excluded from the denominator, and the number of distinct sources and the
	688	+largest single share are reported with the index. Because arXiv is a single
	689	+OpenAlex source, growth in preprints mechanically raises $\mathrm{HHI}_y$ even
	690	+when the journal and conference landscape becomes more dispersed; this is the
	691	+reason the protocol's H5 predicts a stable or falling HHI only ``because of
	692	+preprint growth'' with the preprint source treated separately, and the tables
	693	+report the index with and without the preprint type.
	694	+
	695	+\subsection{Collaboration, countries and institutions}
	696	+
	697	+Team size is summarised by the mean and median number of authors per work and
	698	+by the shares of single-author works and of works with at least five or ten
	699	+authors. Two counting conventions are used for countries and institutions,
	700	+because they answer different questions \citep{waltman2016review}. Under full
	701	+counting a work contributes one unit to every country among its authors'
	702	+affiliations; under fractional counting it contributes $1/\|K(w)\|$ to each of
	703	+its $\|K(w)\|$ distinct countries, so that the fractional totals sum to the
	704	+number of works with at least one country. The same pair of conventions is
	705	+applied to institutions, with the fraction taken over the distinct institutions
	706	+of the work. The international collaboration share of year $y$ is
	707	+\begin{equation}
	708	+ \mathrm{ICS}_y \;=\; \frac{\|\{w : y(w)=y,\ \|K(w)\| \ge 2\}\|}{\|\{w : y(w)=y,\ \|K(w)\| \ge 1\}\|},
	709	+\end{equation}
	710	+that is, among works with at least one resolved country, the share with authors
	711	+from two or more countries; works without affiliation data are excluded from
	712	+the denominator and their share is reported, because affiliation completeness
	713	+in OpenAlex varies by source and year and a change in completeness would
	714	+otherwise masquerade as a change in collaboration. The industry share is the
	715	+share of works with at least one affiliation of OpenAlex institution type
	716	+\textit{company} among works with at least one typed affiliation. Country-pair
	717	+and institution-level tables apply minimum-support thresholds (fifty joint
	718	+works for a country pair, two hundred works for an institution within a period)
	719	+that keep the released tables to the robustly estimated part of the
	720	+distribution; these thresholds are reporting choices and are stated in the
	721	+table notes.
	722	+
	723	+\subsection{Diversity indices}
	724	+\label{sec:m-diversity}
	725	+
	726	+Interdisciplinarity is measured on the distribution of a work's references, or
	727	+of a category's citations, over the categories of the ontology. Let
	728	+$p = (p_1, \dots, p_K)$ be such a distribution and $\delta_{ij} \in [0,1]$ a
	729	+dissimilarity between categories $i$ and $j$. The Rao--Stirling index
	730	+\citep{rao1982diversity, stirling2007diversity}
	731	+\begin{equation}
	732	+ \Delta_{\mathrm{RS}}(p) \;=\; \sum_{i \ne j} p_i\, p_j\, \delta_{ij}
	733	+ \label{eq:rao-stirling}
	734	+\end{equation}
	735	+is the expected dissimilarity between two categories drawn at random from $p$;
	736	+it is zero when all mass sits on one category and increases with variety (the
	737	+number of categories present), balance (the evenness of $p$) and disparity (the
	738	+dissimilarity of the categories involved), the three components that
	739	+\citeauthor{stirling2007diversity} argues any diversity measure should respond
	740	+to. Its use in scientometrics follows \citet{rafols2010diversity}. Its known
	741	+weakness is that it saturates: for very even distributions over many
	742	+categories it approaches the mean dissimilarity and no longer discriminates
	743	+\citep{zhang2016diversity}. The Leinster--Cobbold family
	744	+\citep{leinster2012diversity} repairs this by defining, for a similarity matrix
	745	+$Z = (z_{ij})$ with $z_{ij} = 1 - \delta_{ij}$ and an order $q \ge 0$,
	746	+\begin{equation}
	747	+ {}^{q}D^{Z}(p) \;=\; \Bigl( \sum_{i : p_i > 0} p_i\, (Zp)_i^{\,q-1} \Bigr)^{1/(1-q)},
	748	+ \qquad (Zp)_i = \sum_j z_{ij}\, p_j ,
	749	+ \label{eq:leinster-cobbold}
	750	+\end{equation}
	751	+with the limit ${}^{1}D^{Z}(p) = \prod_i (Zp)_i^{-p_i}$ at $q = 1$. The quantity
	752	+$(Zp)_i$ is the ``ordinariness'' of category $i$ under $p$: how much of the
	753	+distribution is similar to $i$. The index is an effective number of
	754	+categories---it equals $K$ for the uniform distribution over $K$ mutually
	755	+dissimilar categories---and the order $q$ sets the weight given to rare
	756	+categories (large $q$ counts only the dominant ones). Following
	757	+\citet{zhang2016diversity} the manuscript reports $q = 2$, which has the
	758	+closed form ${}^{2}D^{Z}(p) = 1 / \sum_{i,j} p_i z_{ij} p_j = 1/(1 - \Delta_{\mathrm{RS}}(p))$
	759	+when $z_{ii} = 1$, so that the two families are monotone transforms of each
	760	+other at this order and differ only in scale and in their behaviour at the
	761	+saturation end. The dissimilarity $\delta_{ij}$ between categories is one
	762	+minus the cosine similarity of the category centroids of
	763	+Section~\ref{sec:m-semantic} in the baseline period, so that the disparity
	764	+component is estimated from the corpus rather than imposed. The current
	765	+pipeline computes the category flow matrices from which $p$ is formed
	766	+(Section~\ref{sec:m-networks}) and the centroid similarities, but the two
	767	+diversity indices are not yet produced by a script; they are defined here
	768	+because the protocol commits to them, and their values in Part IX are
	769	+\pending{Rao--Stirling and Leinster--Cobbold indices per category and period, not yet computed}
	770	+until the computation is added and audited.
	771	+
556	772

spb/wp44_uqo

WP44: bibliography merge script (599 entries, 15 duplicate keys, 0 DOI conflicts)

Theme