Pre-specified Protocol: Bibliometric Science Mapping of Fuzzy Neural Networks (2020-2026)

Abstract

This deposit is the pre-specified protocol for a bibliometric science mapping of Fuzzy Neural Networks (FNN) and neuro-fuzzy architectures published between 2020 and 2026. It records the study design, the corpus construction procedure, and the analysis plan before any results are generated, so that a reader can check which parts of the analysis were fixed in advance and which were not. The corpus comprises 19,290 Scopus records retrieved on 2026-09-06 through an institutional subscription, using a locked query library of three term blocks (43 exact phrases plus the acronym ANFIS) over TITLE-ABS-KEY with the window PUBYEAR 2020-2026. Collection was manual, one export per publication year, with Excel truncation disabled and no language or document-type filter. The seven files show zero delta against the screen counts and zero duplicate EIDs, and the SHA-256 of each is recorded in the deposited export log. The single-database design is justified by measurement rather than by convenience. Web of Science Core Collection was queried with the identical library, window, and equivalent field scope before it was ruled out: it returned 9,443 records against the Scopus union of 19,290, covering 49% of the corpus, with the gap concentrated in conference literature and regional journals. Because co-citation and bibliographic coupling require internally consistent reference formatting, Scopus is the sole source of records. OpenAlex is used only for cross-verification of coverage (93.51% by DOI) and for recovery of missing first-author country attribution; no record enters the corpus from OpenAlex. The analysis plan is pre-specified in full: descriptive mapping, intellectual structure via co-citation and bibliographic coupling, conceptual structure via keyword co-occurrence and thematic evolution, social structure via co-authorship networks, and a pre-declared test of whether the field is dominated by applied ANFIS work rather than methodological contributions. Inclusion thresholds, normalization, clustering, seed handling, the three-way treatment of citation counts, and the sensitivity analyses that recompute every network at adjacent thresholds are all fixed here in advance, as are the known limitations and the items that are explicitly not pre-specified. No bibliometric analysis had been executed at the time of this deposit. Corpus construction is complete and verified; the analysis has not been run. Every corpus figure stated in the protocol is verified against artifacts on disk and traceable to a dated provenance report included in this package. The bibliographic records themselves are not redistributed here, as they are subject to Elsevier's terms of use. The deposit contains the protocol, the locked query library, the provenance and audit reports, the Python and R environment locks, and the corpus construction and audit scripts. The four core provenance reports, the export log, the corpus audit, the R ingestion log, and the OpenAlex verification, are deposited in both Portuguese and English, as .md and .en.md, with the Portuguese file as the record of authority in each pair. The query, the counts, and the file checksums are sufficient to reproduce the corpus for anyone with Scopus access. See LICENSE_NOTE.md for the licensing of each component. Version 1.11 corrects three counts and adds the context missing from one reported index. In section 10, the count of resolved affiliations read 47,744, the figure under an earlier layer of 133 rules, and now reads 47,784 under the adopted layer of 155 rules, with its base stated. The adjusted Rand index between attributions, reported as 0.1503 without naming the layer measured, is now reported against both layers, 0.1503 and 0.7603, together with the seed-variation measurement that places the second inside the range of algorithmic noise. In section 13, the report count read eleven and now reads fifteen. In section 11, the lockfile count read 124 and now reads 127, the number of entries in every version of the lockfile. Section 11.7 gains the seed-variation procedure, executed after version 1.10 was deposited and declared as such; subsections 11.1 to 11.6 are identical character for character. No corpus figure, search term, inclusion criterion, or analysis decision differs from version 1.10. The superseded values are recorded in section 13 next to the corrected ones. The seed-stability report, the lockfile verification report, and the R script behind the seed measurement are added to this deposit. Versions 1.6 to 1.10 remain accessible under their own DOIs.