Primary academic source: network graph modularity as knowledge-gap signal

Research question: What is the primary academic source for the claim that network graph modularity scores can serve as a knowledge-gap signal?

Answer in brief: The claim, as commonly stated, is imprecise — the actual two-paper foundation is (1) Newman & Girvan (2004), which defines modularity Q, and (2) Paranyushkin (2019, WWW '19), which is the primary peer-reviewed source applying modularity-based community detection to text/knowledge networks and framing the resulting inter-cluster "structural gaps" as a knowledge-gap signal. No single paper makes the condensed claim "modularity scores = gap signal" verbatim; the elision is downstream of both papers.

Claim 1: The primary peer-reviewed source for knowledge-gap detection via community structure in text/knowledge networks is Paranyushkin (2019)

Claim type: Historical (which peer-reviewed work first grounds this claim) → Tier 3–4 acceptable; Tier 1 source available ✓

Dmitry Paranyushkin's paper "InfraNodus: Generating Insight Using Text Network Analysis," presented at The Web Conference (WWW '19), ACM, May 2019, San Francisco, is the primary peer-reviewed source grounding the practice of using modularity-based community detection on text/knowledge graphs to surface structural gaps as knowledge-gap signals. The paper introduces InfraNodus as a system that "represents text as a network where concepts are nodes and their co-occurrences are connections" and applies algorithms from network science to "highlight the most influential concepts, topical clusters, and structural gaps between them."

• source_url: https://doi.org/10.1145/3308558.3314123
• source_author: Dmitry Paranyushkin
• source_date: 2019-05-13
• source_tier: 1
• Exact quote grounding: The full-text PDF (https://noduslabs.com/wp-content/uploads/2019/06/InfraNodus-Paranyushkin-WWW19-Conference.pdf) is confirmed to exist but was inaccessible to text extraction during this run. Quotes below are attributed to the paper via InfraNodus documentation and must be verified against the paper directly before promotion: "We use community detection algorithm based on the modularity measure to cluster the nodes into topical groups" and "identifying structural gaps in discourse." These are marked [unverified-quote — needs direct paper verification] until confirmed against the PDF.

Claim 2: The gap mechanism is a two-step process — community detection first, structural-gap detection second; the modularity score Q is not itself the gap signal

Claim type: Specific technical mechanism → Tier 1–2 required

The "knowledge-gap signal" attributed to modularity involves two distinct steps:

1. Step 1 — Community detection: A modularity-maximisation algorithm (specifically the Louvain method, per InfraNodus) partitions the graph into topical clusters. The modularity score Q is used here only as the optimisation objective for finding clusters — it is not itself the gap signal.

2. Step 2 — Structural-gap detection: A separate algorithm identifies pairs of clusters that are "large and influential but barely connected to each other." These inter-cluster absences — not the Q score — constitute the knowledge-gap signal.

The condensed phrase "modularity scores as knowledge-gap signal" elides this two-step structure. What is actually true: high-Q (strongly modular) graphs exhibit well-separated clusters, and the structural gaps between those clusters are the signal of knowledge territory not yet bridged.

• source_url: https://doi.org/10.1145/3308558.3314123
• source_author: Dmitry Paranyushkin
• source_date: 2019-05-13
• source_tier: 1
• Quote status: [unverified-mechanism — needs direct paper verification] Quote attribution is via InfraNodus product documentation (Tier 3), not confirmed from the ACM paper text directly. Mechanism description is consistent across all InfraNodus materials and is attributed to the Paranyushkin 2019 paper throughout.

Claim 3: The modularity measure Q was introduced by Newman & Girvan (2004) as a quantitative measure of community structure strength

Claim type: Historical/definitional → Tier 3–4 acceptable; Tier 1 source available ✓

M. E. J. Newman and M. Girvan introduced the modularity measure in "Finding and Evaluating Community Structure in Networks," Physical Review E 69, 026113 (2004). The paper defines modularity Q as a measure that compares the observed density of intra-community edges to the density expected under a random null model, providing "an objective metric for choosing the number of communities into which a network should be divided."

• source_url: https://arxiv.org/abs/cond-mat/0308217 (preprint; published version DOI: https://doi.org/10.1103/PhysRevE.69.026113)
• source_author: M. E. J. Newman, M. Girvan
• source_date: 2004
• source_tier: 1
• Exact quoted phrase (from abstract): "a measure for the strength of the community structure found by our algorithms, which gives us an objective metric for choosing the number of communities into which a network should be divided"

The word "modularity" does not appear in the abstract itself — it uses "strength of community structure" — but the measure introduced in the paper body became the canonical definition of Q in network science.

Claim 4: No single paper makes the exact claim "modularity scores serve as a knowledge-gap signal" — this is a downstream synthesis

Claim type: Historical → Tier 3–4 acceptable for uncontested synthesis observation

After systematic web search across arXiv, ACM, Semantic Scholar, and InfraNodus documentation (2026-07-01), no peer-reviewed source was found that makes the verbatim or direct claim that "network graph modularity scores can serve as a knowledge-gap signal." The claim as stated in the parent note appears to be a downstream compression of:

• Newman & Girvan (2004): modularity Q measures community structure strength
• Paranyushkin (2019): community-structured knowledge networks can be mined for structural gaps as insight/knowledge-gap signals

The InfraNodus product and its Obsidian plugin (InfraNodus AI Graph View) operationalise this synthesis in the PKM domain, citing Paranyushkin (2019) as the academic grounding.

• source_url: https://infranodus.com/docs/network-analysis (Tier 3, InfraNodus product documentation)
• source_author: Nodus Labs / Dmitry Paranyushkin
• source_date: undated (retrieved 2026-07-01)
• source_tier: 3
• Note: This is a negative finding (no single primary source) supported by Tier 1–2 search of academic databases. Record this as definitional/historical — no quantitative or specific mechanism claim rests on this Tier 3 source alone.

Further leads

• Blondel et al. (2008), "Fast unfolding of communities in large networks," Journal of Statistical Mechanics: Theory and Experiment — the Louvain algorithm paper that Paranyushkin's system builds on for the modularity-maximisation step; may need its own claim-note if the Louvain mechanism is ever the load-bearing claim. (https://arxiv.org/abs/0803.0476)
• Paranyushkin (2011), "Identifying the Pathways for Meaning Circulation Using Text Network Analysis," Nodus Labs — earlier/pre-peer-reviewed precursor to the 2019 WWW paper; Tier 3-4 (company report, not peer-reviewed journal). Semantic Scholar: https://www.semanticscholar.org/paper/Identifying-the-Pathways-for-Meaning-Circulation-Paranyushkin/97ba9f29c55ccbdd963e01f98cf17e73998f0f7d
• InfraNodus Obsidian plugin documentation mentions "structural gaps — pairs of clusters that are large and influential but barely connected" — clearest operational definition of the gap signal; warrants a dedicated atomic note on the structural-gap concept. (https://noduslabs.com/featured/obsidian-3d-graph-view-plugin-with-network-science-insights/)
• Paranyushkin WWW '19 PDF is confirmed to exist at https://noduslabs.com/wp-content/uploads/2019/06/InfraNodus-Paranyushkin-WWW19-Conference.pdf — a future run should fetch and extract exact quotes from the full paper text to verify the [unverified-quote] flags in Claims 1 and 2 above.
• Abu-Rasheed et al. (2025), "LLM-Assisted Knowledge Graph Completion for Curriculum and Domain Modelling in Personalized Higher Education Recommendations" (arXiv:2501.12300) — uses graph modularity as a validation metric for curriculum knowledge graphs; potentially a second primary source for this application domain.