3

I am trying to find potential clusters or groups of nodes (forum messages, in this case).

In the current data, each node (message) has been tentatively grouped together with n other messages, and that group given a name. So, we know that msg ID 1 has been seen together with msg ID 3, and 7, say.

I am currently using that information to construct an edge list (if they have been grouped together, an edge exists), and then using walktrap community to produce a dendrogram.

Are there any other ways to tease out groups or clusters, given the edge list? (I am using R, but pointers to anything would be helpful).

Thanks for your time!

4

3 回答 3

3

我对R不太了解，但是...

这篇概述论文非常详细地讨论了图聚类。您可能还对马尔可夫聚类(MCL) 算法感兴趣。

于 2012-01-10T19:03:24.153 回答
1

对于 RI 中的网络分析，建议使用 igraph。你已经在使用它了吗？它允许使用不同的聚类方法构建树状图。然后使用 community.to.membership() 切割树状图。MCL 算法（还）不能从 igraph 中获得，但它可以很容易地从命令行运行。

于 2012-01-11T15:36:00.303 回答
1

一种常见的包罗万象的方法似乎是为数据构建一个邻接矩阵，然后使用矩阵乘法来构建一种传递闭包，然后例如计算逆矩阵来识别“原因”，或 PCA/ICA 来识别组.

于 2012-01-15T11:38:37.357 回答