1

I am using R and the IGraph package. my approach is a community detection algorithm based on the eigenvectors centrality . To assess the results of my algorithm, I will use the following evaluation metrics: NMI (Normalized Mutual Information), Purity, ARI (Adjusted Rand Index). but these parameters require the comparison of two communities structures. 

compare(comm1, comm2, method = c("vi", "nmi","split.join", "rand","adjusted.rand"))

Can you offer me one or two algorithms with which I could compare my own?

And how can i get the Purity metric?

Thanks you

deadfish
  • 11,996
  • 12
  • 87
  • 136
Sasa88
  • 327
  • 1
  • 3
  • 15
  • please take a look at this [question](http://stackoverflow.com/questions/9471906/what-are-the-differences-between-community-detection-algorithms-in-igraph) , the algorithm listed in there and also igraph manual as well. – academic.user Mar 25 '15 at 13:45
  • 1
    Note that when evaluating your algorithm, it is best to compare the obtained community structures with some kind of a "ground truth" and not the result of some other algorithm - comparing your result with another algorithm would only tell you whether your algorithm agrees with the other one, and not whether your algorithm agrees with the ground truth that should have been discovered from the data. – Tamás Mar 25 '15 at 19:16
  • @Tamás, effectively, i find in litterature that it's better to compare your algorithm results with ground truth community structure. but i don't know how can i do it. where can i get the real community structure of for example "zachary Karaté club" in **R**. Help me please. There is a method in R with which i can get the real community structure?? – Sasa88 Mar 26 '15 at 10:04
  • If there were such a method, you wouldn't need community detection algorithms at all ;) You need a data set for which the ground truth is specified _a priori_. For what it's worth, the Zachary karate club dataset is one such dataset; download the data from here and check the vertex attribute named `Faction`: http://nexus.igraph.org/api/dataset_info?id=1&format=html – Tamás Mar 26 '15 at 10:39
  • @Tamás. It works for zachary. but not for other datasets. i can't get the real community structure (Faction) for other datasets( dolphins social network, and Network of American college football games ). Any help please.?? – Sasa88 Mar 26 '15 at 16:59
  • The info page for these datasets on Nexus tells you where to look for the ground truth. Incase of the college football dataset, the vertex attribute named `Conference` is the ground truth: http://nexus.igraph.org/api/dataset_info?id=11&format=html . I'm afraid that there is no such attribute in the dataset for the dolphin social network. – Tamás Mar 28 '15 at 20:51
  • @Tamás. I'm very grateful for your help. Effectively, there is no attribute for ground truth community structure for dolphin social network. – Sasa88 Mar 29 '15 at 19:47

0 Answers0