Citation, validation, methods & ethics
Biblio Infinity v1.0 (beta) · 11 June 2026 · this page is the canonical reference for citing the tool and auditing how it computes.
📌 How to cite Biblio Infinity
APA (7th ed.)
BibTeX
RIS (EndNote / Mendeley / Zotero)
License: MIT (full text) · Cite all versions (concept DOI): 10.5281/zenodo.20648810 · v1.0 archive DOI: 10.5281/zenodo.20648811 · archived on Zenodo. A software-paper preprint is on SSRN (abstract 6922358, under SSRN review). Machine-readable metadata: CITATION.cff. Author ORCID: 0000-0001-7682-2244.
✅ Validation — 82 parity checks against reference implementations
Biblio Infinity's metric and network routines are validated against independent reference implementations. The audit has three layers: (1) the exact functions served to your browser are extracted verbatim from the live page and hash-locked (a sync guard fails the suite if the live code ever differs from the validated code); (2) the extracted functions run on a 90-document reference corpus and must match values computed independently in Python (networkx, python-louvain, textbook formulas); (3) clustering quality is cross-checked between languages. The full harness is downloadable below.
| Module | Validated against | Checks | Agreement |
|---|---|---|---|
| Scopus CSV parsing (records, authors, keywords, references, citations, years) | known-design corpus (90 docs) | 6 | exact |
| Corpus statistics (mean citations, h-index, g-index, international co-authorship, top source/country) | manual Python (textbook definitions: Hirsch 2005; Egghe 2006) | 9 | exact / ≤ 1e-9 |
| Keyword co-occurrence graph (item occurrence + full-counting link weights, minOcc filter) | itertools pair counting (Python) | ~20 | exact |
| Association-strength normalization (van Eck & Waltman 2009) | formula audit on every link + Python cross-check | 2 | ≤ 1e-12 |
| Betweenness centrality (Brandes 2001) | networkx betweenness_centrality (normalized) | 1 (max abs diff over all nodes) | ≤ 1e-9 |
| Louvain clustering + modularity Q (Blondel et al. 2008; Newman 2006) | networkx modularity on the same partition + python-louvain best-partition quality band | 4 | Q cross-language ≤ 5e-4 (display rounding); quality ≥ best − 0.05 |
| Network-level metrics (density, mean degree, components, clustering coefficient, mean path length, diameter) | networkx | 8 | exact / ≤ 1e-9 |
| Reference co-citation & bibliographic coupling edge weights (Small 1973; Kessler 1963) | itertools/set counting (Python) | ~10 | exact |
| RPYS — reference publication year spectroscopy (Marx et al. 2014) | independent median-deviation implementation (Python) | 3 (incl. all peak years) | exact |
| Bradford's Law zones & Lotka's Law fit (Bradford 1934; Lotka 1926) | manual Python (cumulative thirds; log-log least squares) | 6 | exact / ≤ 1e-9 |
| Kleinberg burst detection (Kleinberg 2003) | independent 2-state Viterbi implementation (Python): burst interval on synthetic spike + flat-series null | 3 | interval exact |
| Reference-string parsing (refKey/refYear, Scopus & WoS formats), country canonicalization | known cases | 5 | exact |
| Cross-language phase (networkx Q of tool partition; python-louvain quality band) | — | 2 | pass |
Total: 82 / 82 checks passing on the v1.0 build, plus the sync guard proving the audited functions are byte-identical to the code served at /biblioinfinity/. The reference corpus is simulated with known structure (3 keyword themes + bridge terms, Bradford-skewed sources, citation classics seeded in references), so recovery of the designed structure is itself part of the audit.
⬇ Download validation package (extractor + corpus + reference generators + tests)
🔬 Methods & algorithms implemented
- Science mapping. Keyword co-occurrence, co-authorship, country collaboration, direct citation, reference co-citation (Small, 1973) and bibliographic coupling (Kessler, 1963) networks; full and fractional counting; association-strength normalization (van Eck & Waltman, 2009); standard map/network file import/export (.map, .net, JSON) for interoperability with other science-mapping software.
- Clustering & layout. Louvain community detection with resolution parameter (Blondel et al., 2008), modularity Q (Newman, 2006), per-cluster silhouette, force-directed layout.
- Structure & dynamics. Brandes (2001) betweenness centrality with pivotal-point flagging (Chen, 2004); Kleinberg (2003) burst detection; sigma novelty (Chen et al., 2010); timeline and overlay views; historiograph of direct citations.
- Classic bibliometrics. Corpus h-index (Hirsch, 2005) and g-index (Egghe, 2006); Bradford's Law zones (Bradford, 1934); Lotka's Law fit (Lotka, 1926); thematic strategic map by Callon centrality × density (Callon et al., 1991); RPYS (Marx et al., 2014); three-fields plot; growth heatmaps; world collaboration map.
- Data handling. Scopus CSV, Web of Science, PubMed, BibTeX and RIS parsing; OpenAlex corpus building; duplicate merging; thesaurus-based term unification — all client-side.
🔁 Reproducibility
Identical inputs and settings reproduce identical maps: clustering uses a fixed random seed, and all counting is deterministic. Figures export as editable SVG and high-resolution PNG with embedded captions; reports export as standalone HTML and multi-sheet Excel with the underlying tables. The validation package lets any researcher re-run the full audit in minutes (bash run_bi_validation.sh).
🛡 Ethics, privacy & AI transparency
- Your data never leave your browser. All parsing and computation are client-side. Nothing is uploaded or stored server-side.
- AI assistance is optional and aggregate-only. If you use AI features (cluster naming, map interpretation, gaps, synthesis), only aggregated cluster terms and counts are sent to the AI service — never your file. Every AI output is labelled as machine-generated with a verify-before-citing note, and AI sections in exported reports carry the same label. Disclose AI assistance where your journal's policy requires it; responsibility rests with the authors.
- Interoperability. The tool reads and writes widely used bibliometric map and network file formats (.map, .net, JSON), so corpora and maps can move between Biblio Infinity and other science-mapping software.
⚠️ Known limitations (v1.0)
- Betweenness centrality is computed for networks up to 1,200 nodes; clustering coefficient and path metrics are sampled on very large graphs (flagged as approximate in the metrics panel).
- Reference-based analyses (co-citation, coupling, RPYS, historiograph) require exports that include reference lists; reference-string matching is heuristic (first author + year) and can merge or split variants.
- Country detection parses affiliation strings heuristically; check the country table before publication.
- Louvain is non-deterministic across resolutions in principle; Biblio Infinity fixes the seed for reproducibility, but different resolutions yield different (valid) partitions.
- AI interpretations are drafts, not findings — verify against the literature.
🗓 Release notes
v1.0 — 11 June 2026. First public release (beta). Full pipeline: import (Scopus/WoS/PubMed/BibTeX/RIS/OpenAlex) → cleaning → statistics dashboard → science maps (co-occurrence, co-authorship, citation, co-citation, coupling) → advanced metrics (betweenness pivotal points, Kleinberg bursts, sigma, silhouette, modularity) → classic laws (Bradford, Lotka), thematic strategic map, RPYS, three-fields plot, world map, historiograph → AI insights → publication-ready figure and report export. 82-check validation audit published with sync guard; MIT license; citation kit. (Internal redesign history: v9 "Infinity" UI, June 2026.)