• December 2013
    M T W T F S S
    « Nov   Jan »
  • Pages

  • Marginalia

  • Accumulations

  • Advertisements

What Is Modularity?


Above, “natural clusters” — a term that needs a definition — in the network of Demo Day 21212, held in October and intended to get state, private and foreign venture capitalists together with Brazilian startups, both academic and non-academic. Along with advertorial-ridden press with conflict of interest problems to promote the event.

Therefore, we should check to see if the categories can be sorted by algorithm into a more or less thematically unified overview, as a precursor to the selection of subnets.

A simple Fruchterman with a heat map provides a sample of the universe of entities involved.


The question is how to find communities performing their distinct functions in this complex but highly coordinated process.

Captura de tela de 2013-11-30 15:11:47

Source: Modularity | Gephi Wiki

A high modularity score indicates sophisticated internal structure.

This structure, often called a community structure, describes how the the network is compartmentalized into sub-networks. These sub-networks (or communities) have been shown to have significant real-world meaning.

Apart from the Pajek drawing tools, I have been alternating between Gephi and yEd in my amateur attempts to detect and identify significant ties in induced subnetworks.


At my level of ability, this is an exercise involving more intuition than anything else — I read a lot of blogs with the aid of some basic contextual SNA, some basic interpretation of structural results learned from Pajek and gephi.

For example, in the network segment shown above,

Dark blue = Government sites (science and tech)

Olive green = Associates of the Instituto Millennium

Lilac: a Globo show on SMEs features content from Endeavor, a school for entrepreneurship

Intuition says the nodes that are lumped together belong together in the graph shown below — an egocentric view of a network focusing on recently formed alliances in the alternative press, such as Carta Capital, Carta Maior, and Jornada of Mexico.

It seems to make sense. Periodismo (red) is a strong hub containing an international selection of  publications and broadcasts, many of them Brazilian: everything from O Globo to a significant concentration of radio stations. We should study these, their ownership and operations.

But how does it work? What accounts for the reported community divisions in the graph shown above, for example?


  1. Light blue is dominated by Argentine and Chilean dailies — with an anomalous appearance by this blog! — and connected to IstoE magazine. Other Chilean and Argentine media are found in the red partition and are connected to it through the Periodismo hub.
  2. Grey is the color of the consultancy 21212, which plans and executes VC-entrepreneur meet-ups and has ties to the Kennedy School at Harvard — a major influence. It also includes digital consultancy Movile, whose role could be further explored.
  3. Red is difficult to read due to its density, though the zoom can be adjusted — It seems to be dominated by Globo print and radio and congeners in Chile and Argentina. What institutional ties do they have, if any?
  4. Deep purple: Advertising houses Ernst & Young and Maquina.

Inspection of “associates” rosters in this subnet indicate how sophisticated and expensve the marketing campaign was:

  1. Grupo Máquina (PR)
  2. 22 Graus (MG)
  3. Bojo Conteúdo (RS)
  4. Dialetto Comunicação Estratégica (SC)
  5. Janela Comunicação (PE)
  7. Grupo RBS
  8. Organizações Globo
  9. Yahoo!
  10. InfoMoneyNow!
  11. Digital BusinessStartup
  12. Catraca
  13. LivreEnox
  14. On Life
  15. MediaTagarela

The question is: What grouping algorithm is there, if any, whose modularity measurement results in finer-grained communities, suitable for grouping?

Captura de tela de 2013-11-30 11:16:49

It looks as though the event will be (was?) sponsored by the privately-held iG group and a project at the MCTI — Ministry of Science and Tech — outsourced to 21212.

Captura de tela de 2013-11-30 11:28:50

Relations among the most central nodes: 21212, Startup iG, and Totvs, Brazil’s acquisitive and rapidly growing enterprise software house.

Captura de tela de 2013-11-30 12:03:38

Working with gephi is very rewarding, though frustrating at times. Its data laboratory is very handy. When data is drawn in yEd, on the other hand, it is much easier to sift and remix. The process requires translating files in the .net format used by Pajek into a CSV and finally into an XP/2003 Excel spreadsheet file. A sample:


Note the handy drawing tools you can use to improve on or expand the current file.

This particular subnetwork originally looked like the following using Pajek Draw tools.


When it comes time to inspect the distribution of nodes in various groupings, yEd has a very useful window.

Captura de tela de 2013-11-30 15:11:47

Groups are identified after the Node ID of the highest centrality degree of the internal structure, which are most often MAN numbers 300 and 120.


I still think there is something to using SNA to understand the structure and function of varoius modes of networked organizations, according to Lethbridge — which immerses us in the theory of the incubator and its supporting structures. Above, an egocentric sketch of structures from the customerv viewpoint. This was back in 2012]. I wrote a note on international VBEs — viritual breeding environments, or incubators — around the same time.

To be continued …

Captura de tela de 2013-11-30 12:03:38

Captura de tela de 2013-11-30 11:28:50

Captura de tela de 2013-12-02 11:18:19 Captura de tela de 2013-12-02 12:01:49 Captura de tela de 2013-12-02 13:54:12Captura de tela de 2013-12-03 11:21:59 Captura de tela de 2013-12-03 11:24:34 Captura de tela de 2013-12-03 11:57:14 Captura de tela de 2013-12-03 11:57:26 CENPEC Links Endeavor364 ImiligStartup Endeavor364 ImiligStartup IMILINtergroups JornadaCartaMaiorIstoE Screenshot from 2013-11-07 12:22:41 StartingConfiguration  toolbox