***** To join INSNA, visit http://www.insna.org ***** Carlos, muchos gracias. (That is, I'm afraid, the end of my Spanish). I have just downloaded the Santo Fortunato article and am looking forward to reading it. I should add to what I have written in other messages that I am continuing to work on a problem that I talked about at the Sunbelt Conference in Hamburg. I have a body of data — networks of award-winning ads and the creators who produced them, sampled at five-year intervals—from an industry I know a good deal about, partly from recent historical and ethnographic research, and partly because I worked for one of Japan's two largest advertising agencies for thirteen years. The teams that created the ads have attributes. These include (1) the agencies that produced the ads and (2) the medium for which each ad was produced. There is plenty of both qualitative and quantitative data to support the proposition that during the period in question TV captured both a larger share of awards and a larger share of advertising budgets, while print advertising declined, and that this was also a period in which the largest agency in the industry also captured a larger share of awards and a larger share of budgets. While attempting to determine the relationship between components and agency or media networks, I became aware of the general problem of relating the output of network analysis algorithms to known facts about the networks in question, like those mentioned above. It turns out that with Pajek it is very simple to determine, for example, the number of creators who work on award-winning ads for two or more agencies during the same year and demonstrate that the networks' giant components are only thin veils over what remains a highly stovepiped industry. As Vlado (Vladimir Batagelej) quipped after my presentation I demonstrated that Pajek can count. But, if all I am doing is shrinking networks around extended partitions, why not see what happens with partitions generated by valued core, Louvain, or VOS or other clustering algorithms....This anyway is the thought driving my curiosity about detailed differences in expectations concerning the output of various methods. Again, thanks for your advice and the link to the Fortunato paper. John On Mon, Aug 5, 2013 at 11:06 PM, Carlos Sanz Rodriguez <[log in to unmask]>wrote: > Of course there a fundamental different between two methods. VOS is base > on betweenness centrality and Louvain is base on modularity. This one > difference of approaches of the problem generate two different views of > your networks. Lovain has a resolution limits and this cause a problem with > isolated nodes, but is very good to approximate to a modular structure of a > network. VOS is a good to resolve a community structure in different level, > but have the problem the many nodes with certain betweenness score are > inside the community in the reality and not outside like frequently the > method said, so in this case Louvain resolve better the structure than VOS > in many case. The decision many times come from how well you know your data > and the behavior of the your communities. In my case, a study metabolic > network and almost all the time the communities correlated with some > function in the metabolism, so the method hat I apply and select to my > 'after analysis', depend on how week the algorithm describe the > organization of my data and function inside the metabolism. > There are a very good an extensive review of a community detection methods > from Santo Fortunato: Physics Reports (2010) Vol. 46 p. 75.In this Review > the VOS methods referred like 'Newman approach'. There are many, many > methods for community detection and there are not a master method, all of > them have a good performance to a some networks and are very bad for > others. These methods that offer Pajek are too popular and using for many > people, but that are not perfect. > Well, I hope that you can found what method is better for you question. > Best, > Carlos > > Carlos Sanz Rodriguez > Departamento de Biología Celular > Universidad Simón Bolívar > Caracas, Apartado 89000 > Telf +58-212-9064219 > > Laboratorio de Dinámica Estocástica > Centro de Fisica, I.V.I.C. > Caracas, Apartado 1020-A > Telf +58-212-5041919 > > El 04/08/2013, a las 21:28, John McCreery <[log in to unmask]> > escribió: > > Two recent runs through part of my data using the Pajek commands for, > first, VOS and, second, Louvain community detection. My question is why the > results using the Louvain method seem lumpier than those using the VOS > method. Note, for example, that cluster 1 in the VOS distribution has 36 > members, while cluster 1 using the Louvain method has 97. Using the VOS > method, cluster 11 has 35 members, while cluster 11 using the Louvain > method has 101. Is there a straightforward explanation why the two > algorithms produce such apparently different results? Or is this just a > random event? > > Thoughts? > > John (data follows) > > (1) > > > > ============================================================================== > 1. VOS Clustering in N1 (637, Res=1.000000, VOS=0.9010200019, NC=54) > > ============================================================================== > Dimension: 637 > The lowest value: 1 > The highest value: 54 > > Frequency distribution of cluster values: > > Cluster Freq Freq% CumFreq CumFreq% Representative > ---------------------------------------------------------------- > 1 36 5.6515 36 5.6515 AD1_01 > 2 23 3.6107 59 9.2622 AD9_01 > 3 67 10.5181 126 19.7802 AD15_01 > 4 30 4.7096 156 24.4898 AD25_01 > 5 31 4.8666 187 29.3564 AD26_01 > 6 42 6.5934 229 35.9498 AD30_01 > 7 41 6.4364 270 42.3862 AD34_01 > 8 24 3.7677 294 46.1538 AD38_01 > 9 42 6.5934 336 52.7473 AD41_01 > 10 3 0.4710 339 53.2182 AD46_01 > 11 35 5.4945 374 58.7127 AD53_01 > 12 19 2.9827 393 61.6954 AD61_01 > 13 48 7.5353 441 69.2308 AD66_01 > 14 37 5.8085 478 75.0392 AD70_01 > 15 45 7.0644 523 82.1036 AD80_01 > 16 7 1.0989 530 83.2025 AD84_01 > 17 2 0.3140 532 83.5165 AD103_01 > 18 24 3.7677 556 87.2841 AD111_01 > 19 1 0.1570 557 87.4411 AD140_01 > 20 13 2.0408 570 89.4819 AD143_01 > 21 2 0.3140 572 89.7959 AD159_01 > 22 1 0.1570 573 89.9529 AD170_01 > 23 4 0.6279 577 90.5808 AD173_01 > 24 1 0.1570 578 90.7378 AD181_01 > 25 1 0.1570 579 90.8948 AD222_01 > 26 22 3.4537 601 94.3485 AD225_01 > 27 1 0.1570 602 94.5055 AD250_01 > 28 3 0.4710 605 94.9765 AD289_01 > 29 1 0.1570 606 95.1334 AD309_01 > 30 1 0.1570 607 95.2904 AD316_01 > 31 2 0.3140 609 95.6044 AD344_01 > 32 2 0.3140 611 95.9184 AD346_01 > 33 1 0.1570 612 96.0754 AD350_01 > 34 2 0.3140 614 96.3893 AD352_01 > 35 2 0.3140 616 96.7033 AD353_01 > 36 1 0.1570 617 96.8603 AD354_01 > 37 1 0.1570 618 97.0173 AD356_01 > 38 1 0.1570 619 97.1743 AD389_01 > 39 1 0.1570 620 97.3312 AD394_01 > 40 2 0.3140 622 97.6452 AD395_01 > 41 1 0.1570 623 97.8022 AD420_01 > 42 1 0.1570 624 97.9592 AD471_01 > 43 2 0.3140 626 98.2732 AD544_01 > 44 1 0.1570 627 98.4301 AD546_01 > 45 1 0.1570 628 98.5871 AD560_01 > 46 1 0.1570 629 98.7441 AD564_01 > 47 1 0.1570 630 98.9011 AD580_01 > 48 1 0.1570 631 99.0581 AD592_01 > 49 1 0.1570 632 99.2151 AD593_01 > 50 1 0.1570 633 99.3721 AD621_01 > 51 1 0.1570 634 99.5290 AD623_01 > 52 1 0.1570 635 99.6860 AD624_01 > 53 1 0.1570 636 99.8430 AD630_01 > 54 1 0.1570 637 100.0000 AD634_01 > ---------------------------------------------------------------- > Sum 637 100.0000 > > (2) > > > > ============================================================================== > 1. Louvain Communities in N4 (637, Res=1.000000, Q=0.601442, NC=53) > > ============================================================================== > Dimension: 637 > The lowest value: 1 > The highest value: 53 > > Frequency distribution of cluster values: > > Cluster Freq Freq% CumFreq CumFreq% Representative > ---------------------------------------------------------------- > 1 97 15.2276 97 15.2276 AD1_01 > 2 68 10.6750 165 25.9027 AD15_01 > 3 48 7.5353 213 33.4380 AD25_01 > 4 45 7.0644 258 40.5024 AD34_01 > 5 23 3.6107 281 44.1130 AD37_01 > 6 17 2.6688 298 46.7818 AD38_01 > 7 3 0.4710 301 47.2527 AD46_01 > 8 58 9.1052 359 56.3579 AD50_01 > 9 33 5.1805 392 61.5385 AD53_01 > 10 11 1.7268 403 63.2653 AD61_01 > 11 101 15.8556 504 79.1209 AD66_01 > 12 7 1.0989 511 80.2198 AD84_01 > 13 2 0.3140 513 80.5338 AD103_01 > 14 14 2.1978 527 82.7316 AD110_01 > 15 25 3.9246 552 86.6562 AD111_01 > 16 28 4.3956 580 91.0518 AD123_01 > 17 1 0.1570 581 91.2088 AD140_01 > 18 2 0.3140 583 91.5228 AD159_01 > 19 1 0.1570 584 91.6797 AD170_01 > 20 4 0.6279 588 92.3077 AD173_01 > 21 1 0.1570 589 92.4647 AD181_01 > 22 1 0.1570 590 92.6217 AD222_01 > 23 4 0.6279 594 93.2496 AD224_01 > 24 1 0.1570 595 93.4066 AD250_01 > 25 3 0.4710 598 93.8776 AD289_01 > 26 1 0.1570 599 94.0345 AD309_01 > 27 1 0.1570 600 94.1915 AD316_01 > 28 3 0.4710 603 94.6625 AD322_01 > 29 2 0.3140 605 94.9765 AD344_01 > 30 1 0.1570 606 95.1334 AD350_01 > 31 2 0.3140 608 95.4474 AD352_01 > 32 2 0.3140 610 95.7614 AD353_01 > 33 1 0.1570 611 95.9184 AD354_01 > 34 1 0.1570 612 96.0754 AD356_01 > 35 1 0.1570 613 96.2323 AD389_01 > 36 1 0.1570 614 96.3893 AD394_01 > 37 2 0.3140 616 96.7033 AD395_01 > 38 1 0.1570 617 96.8603 AD420_01 > 39 2 0.3140 619 97.1743 AD447_01 > 40 4 0.6279 623 97.8022 AD462_01 > 41 1 0.1570 624 97.9592 AD471_01 > 42 2 0.3140 626 98.2732 AD544_01 > 43 1 0.1570 627 98.4301 AD546_01 > 44 1 0.1570 628 98.5871 AD560_01 > 45 1 0.1570 629 98.7441 AD564_01 > 46 1 0.1570 630 98.9011 AD580_01 > 47 1 0.1570 631 99.0581 AD592_01 > 48 1 0.1570 632 99.2151 AD593_01 > 49 1 0.1570 633 99.3721 AD621_01 > 50 1 0.1570 634 99.5290 AD623_01 > 51 1 0.1570 635 99.6860 AD624_01 > 52 1 0.1570 636 99.8430 AD630_01 > 53 1 0.1570 637 100.0000 AD634_01 > ---------------------------------------------------------------- > Sum 637 100.0000 > > -- > John McCreery > The Word Works, Ltd., Yokohama, JAPAN > Tel. +81-45-314-9324 > [log in to unmask] > http://www.wordworks.jp/ > _______________________________________________ > Pajek mailing list > [log in to unmask] > http://list.fmf.uni-lj.si/cgi-bin/mailman/listinfo/pajek > > > -- John McCreery The Word Works, Ltd., Yokohama, JAPAN Tel. +81-45-314-9324 [log in to unmask] http://www.wordworks.jp/ _____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.