Hello,
I'm running spinglass community detection algorithm on my dataset for which i have some kind of "ground truth". I run Danon's comparison of the clustering to see the "correlation" between the two clusterings. I compute a value of about 0.43. As much as I know that the range is between 0 (independence) and 1 (identical), it is not intuitive to interpret 0.43. As the data is not really ground truth (hence we're not really computing "accuracy"), and the data is real and social behavior (hence you'd not expect 1.0 as with ad-hoc networks), I do expect that 0.43 is a good value, but I want to be able to show it by "comparison".
Do you have some references to papers that do run Danon on clusterings extracted with community detection algorithms such as spinglass (or whatever for what it's worth), on real social networks (FB, friendster, twitter dataset and whatnot)? This would allow me to estimate an "upper bound" compared to the ad-hoc networks which are usually used to validate community detection algorithms (also through Danon).
Thanks.
Best,
Claudio
--
Claudio Martella
address@hidden