Practical guide · Web intelligence

PageRank and web ecosystem centrality: map without overinterpreting

Understand PageRank, betweenness, harmonic centrality and Louvain communities in a bounded web network, with the limits needed for a defensible reading.

Key points

What to remember

  1. 01

    Centrality depends on the constructed graph.

  2. 02

    Each metric answers a different question.

  3. 03

    Louvain yields structural communities, not real organisations.

  4. 04

    Centrality, quality and real influence remain separate.

01

Bound the observed web ecosystem

A map starts with a question, seed sites, inclusion and exclusion rules, depth and date. It represents the arena actually collected rather than the whole web or market.

The denominator remains visible: sites actually crawled, domains only referenced, retained link types and missing content. Changing these choices changes the graph and its centralities.

  • Investigation question
  • Justified seed sites
  • Collection and exclusion rules
  • Depth, volume and date
  • Stable node and edge granularity
02

Build a directed graph before ranking

In a web arena, a node may represent a domain and an edge A → B may mean that A cites B. Co-citations, edge direction and any weight convention must be explained before reading a rank.

Centrality is never independent of the model. Aggregating by domain rather than page, weighting a relationship or expanding collection can change the order even when no real actor has changed.

ElementControl questionEffect on interpretation
NodePage, domain or organisation?Changes comparison granularity
EdgeCitation, co-citation or inferred relationship?Changes network meaning
DirectionWho cites whom?Conditions directed metrics
WeightCount, frequency or distance?Conditions calculation and meaning
03

Choose the method for the question

PageRank recursively distributes importance through incoming-link structure. Betweenness measures the share of shortest paths passing through a node. Harmonic centrality sums reciprocal distances and remains useful in graphs that are not fully connected.

Louvain heuristically searches for a partition that increases modularity, then aggregates communities and repeats. The result depends on the graph, weights, resolution and processing order; it describes structural groups, not real organisations, segments or coalitions.

MethodUseful questionForbidden conclusion
PageRankWhich nodes receive weight from nodes that are themselves central?This is Google’s global PageRank
BetweennessWhich nodes sit on many structural paths?They are indispensable in the real world
Harmonic centralityWhich nodes remain close to the reachable graph?They spread information faster in reality
LouvainWhich node groups are more densely connected under modularity?They are real organisations or segments
04

Read scores without inventing authority

Centrality is relative to the corpus and calculation convention. It does not directly measure truth, editorial quality, commercial reputation, traffic or decision power.

A missing result means the measure was not calculated or did not yield a usable value; it must not be replaced with zero. Comparisons over time require the same granularity, rules and documented parameters.

  • Centrality ≠ quality
  • Rank ≠ causality
  • Community ≠ real organisation
  • No link ≠ no relationship
  • Not calculated ≠ zero
05

Turn rank into a verification programme

Central nodes, bridges and poorly covered areas help select what to verify: dependence on a few sources, a missing actor, a passage between communities or an under-documented topic. They direct the investigation rather than settle it.

For SEO and GEO, a gap weighted by arena centrality can prioritise editorial research. It does not guarantee ranking, traffic or citation by a search engine or assistant.

  1. 01

    1. Detect

    Identify a rank, bridge or low-coverage area.

  2. 02

    2. Return to the corpus

    Retrieve the links, documents and rules producing the signal.

  3. 03

    3. Seek contradiction

    Add independent sources and missing actors.

  4. 04

    4. Decide

    Record what the signal changes, does not change and when to review it.

06

What Nexus supports

Nexus is an Innovatio workspace outside the seven-workspace Innovatio Decision Suite. It structures bounded web investigations, calculates PageRank and betweenness, and retains provenance in JSON or GEXF exports. Harmonic centrality and Louvain remain complementary methodological references here and are not presented as calculations performed by Nexus.

The scope depends on the corpus, collection rules and analysis date. It assumes neither unlimited collection nor systematic Atlas integration. Calculations guide human investigation; they do not make an autonomous decision.

FAQ

Frequently asked questions

Does PageRank measure an actor’s real authority?

No. It measures a relative position in the constructed graph. Real authority needs complementary evidence.

Is this the PageRank used by Google?

No. This page describes a measure calculated in the Nexus arena, not Google’s global proprietary ranking system.

Why can PageRank and betweenness identify different nodes?

PageRank reads recursive incoming-link distribution, while betweenness reads paths passing through a node.

What does a Louvain community mean?

A structural group obtained through heuristic modularity optimisation in the constructed graph, not a real organisation, segment or coordination.

Is Nexus part of the Decision Suite?

No. Nexus is a separate Innovatio workspace outside the Suite of Atlas, Grid, Plot, Mix, Calc, Map and Pilot.

SRC

Full sources