[SOLVED] Partitioning. community finding

[mikebites]

I feed Gephi a graph of around 1152 nodes. Then I calculate the modularity and partition it based on modularity (with node coloring). What I see is nodes with different colors are mixed and scattered all over the graph. My question is- is there any way to bring the group the nodes of each color together?

[admin]

Hi,

Could you take a screenshot of Gephi? Most layouts make visual clusters fitting with this metric.

[mikebites]

Screenshot attached.

Also, I have another query, say a graph has nodes A,B,C, D, E, F, G, H at time t1- where A, B, C is in Partition 1, D, E, F is in partition 2 and G,H is in partition 3. Now on this data, I do the modularity calculation and partitioning and say the colors of the partitions are green, yellow and blue respectively. In time t2 (another graph for t2 is fed) two more nodes join I, J. I becomes a part of partition 1 and J creates a new partition 4 along with C. So at t2, partition 1 <A, B, I>, partition 2 <D, E, F>, partition 3 <G, H>, partition 4 <C, J>. So will the partitions in this case will show green, yellow, blue and a new color or will they be assigned in a new random arrangement. How can I make sure that there is coherence of color assignement of communities? Also If I want to merge partition 1 and partition 4 and assign all the nodes the color that nodes of partition 1 have, is it possible?

[admin]

I agree about the need of coherence, but we didn't tested it so far. New nodes are in black by default. Let us informed about your experiment, and we will correct the software as expected!

If I want to merge partition 1 and partition 4 and assign all the nodes the color that nodes of partition 1 have, is it possible?

This feature is missing, I edited the Ideas list.

[mikebites]

Gephi is a nice tool and it would be nice to see it grow even better. I use it for my scientific work and suggest it to others. It possibly is one of the best freely available tools to visualize graphs and analyze it at the same time.

Another important thing in partitioning may be specifying number of partitions, like what if I want to see best five clusters. Four best clusters and rest of the nodes forming another. Also specify number of nodes in one partition may be another feature you could think of implementing. Recently it has been seen that in social networks, community size is more or less constant (around 100). So what happens if some one wants to see partitioning where size of partitioning do not exceed 100? If you could implement this feature that would be interesting.

[admin]

This feature concerns community detection, performed by statistical algorithms. Gephi has currently one algorithm, Modularity, which provides no option (for the moment). Our approach, in Gephi team, is to develop generic functionalities that makes specific ones easy to add, like layouts, stats or file I/O. If a requested feature needs core development, we will do it like "partition merging". If it only needs the use of existing APIs like the Statistics API, we let someone of the community implementing it, helping him to achieve this plugin. That's the deal to be quick

So if you know a clustering algorithm fitting with this social networks property, we'd glad to help you creating your first Gephi plugin. A Dev Quick Start is available here, and we'll write soon a HowTo for metrics, like the layout's one.

Would you like to try?

Note that you'll have the opportunity to publish it on the Gephi Plugins Platform, making it available to others through the Update Center of each Gephi instances (4500 DL so far ^^).

[mikebites]

Hmm....Thanks for the pointers. I'll have a look at them and if I feel that it could be of help to me, obviously I'll try them out. Cheers.