Consider the diversity of evaluation metrics, that different algorithms detect distinct communities in the same network and that the ground truth does not correspond to the topological structure. In this context, how to properly evaluate the quality of a community structure?
Dear Jeancarlo,
that is a very good question. I think that Danon et. al's work (https://arxiv.org/pdf/cond-mat/0505245.pdf) provides a nice comparison among several community detection by performing a sensitivity analysis on ad hoc networks.
Best Regards.
We propose a more consilient solution to this problem. The experimental results can be found in: Conference Paper A Multi-Strategy Approach to Overcoming Bias in Community De...
There is the concept of entropy using tens of different types of entropies. You can try to find research using them or to develop your own approach.
In bio-signal classification & prediction, we use many entropy measures like
* approx approximation entropy
* sample entropy
* Kolmogorov entropy
* information entropy (Shannon e.)
* And many others
They enable to access the hidden information in signals but they, or similar ones can be used in assessing the entropy (the level of organization) in parts of networks too.
Actually, it sounds like an exciting approach to research. But sorry, I have no idea whether someone already used it or not. When there is enough time, it would be interesting even for myself to carry on such exciting research. :-)
You can find some sort of starting information & citations in my review paper on Complex Systems in medicine, which is freely available here on RG for starting information. In the future (half of the year or so, I will publish a paper dealing with the prediction of biosignals using entropy(ies).
.
we optimize an MDL-derived criterion as described in
http://www.marc-boulle.fr/publications/GuigouresEtAlADAC18.pdf
or similarly
Article Bayesian stochastic blockmodeling
.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs