R igraph manual pages

Community structure detecting based on the leading eigenvector of the community matrix

Description

This function tries to find densely connected subgraphs in a graph by calculating the leading non-negative eigenvector of the modularity matrix of the graph.

Usage

cluster_leading_eigen(graph, steps = -1, weights = NULL,
  start = NULL, options = arpack_defaults, callback = NULL,
  extra = NULL, env = parent.frame())

Arguments

graph	The input graph. Should be undirected as the method needs a symmetric matrix.
steps	The number of steps to take, this is actually the number of tries to make a step. It is not a particularly useful parameter.
weights	An optional weight vector. The ‘weight’ edge attribute is used if present. Supply ‘NA’ here if you want to ignore the ‘weight’ edge attribute. Larger edge weights correspond to stronger connections between vertices.
start	NULL, or a numeric membership vector, giving the start configuration of the algorithm.
options	A named list to override some ARPACK options.
callback	If not NULL, then it must be callback function. This is called after each iteration, after calculating the leading eigenvector of the modularity matrix. See details below.
extra	Additional argument to supply to the callback function.
env	The environment in which the callback function is evaluated.

Details

The function documented in these section implements the ‘leading eigenvector’ method developed by Mark Newman, see the reference below.

The heart of the method is the definition of the modularity matrix, B, which is B=A-P, A being the adjacency matrix of the (undirected) network, and P contains the probability that certain edges are present according to the ‘configuration model’. In other words, a P[i,j] element of P is the probability that there is an edge between vertices i and j in a random network in which the degrees of all vertices are the same as in the input graph.

The leading eigenvector method works by calculating the eigenvector of the modularity matrix for the largest positive eigenvalue and then separating vertices into two community based on the sign of the corresponding element in the eigenvector. If all elements in the eigenvector are of the same sign that means that the network has no underlying comuunity structure. Check Newman's paper to understand why this is a good method for detecting community structure.

Value

cluster_leading_eigen returns a named list with the following members:

membership	The membership vector at the end of the algorithm, when no more splits are possible.
merges	The merges matrix starting from the state described by the membership member. This is a two-column matrix and each line describes a merge of two communities, the first line is the first merge and it creates community ‘N’, N is the number of initial communities in the graph, the second line creates community N+1, etc.
options	Information about the underlying ARPACK computation, see arpack for details.

Callback functions

The callback argument can be used to supply a function that is called after each eigenvector calculation. The following arguments are supplied to this function:

membership

The actual membership vector, with zero-based indexing.

community

The community that the algorithm just tried to split, community numbering starts with zero here.

value

The eigenvalue belonging to the leading eigenvector the algorithm just found.

vector

The leading eigenvector the algorithm just found.

multiplier

An R function that can be used to multiple the actual modularity matrix with an arbitrary vector. Supply the vector as an argument to perform this multiplication. This function can be used with ARPACK.

extra

The extra argument that was passed to cluster_leading_eigen.

The callback function should return a scalar number. If this number is non-zero, then the clustering is terminated.

Author(s)

Gabor Csardi csardi.gabor@gmail.com

References

MEJ Newman: Finding community structure using the eigenvectors of matrices, Physical Review E 74 036104, 2006.

See Also

modularity, cluster_walktrap, cluster_edge_betweenness, cluster_fast_greedy, as.dendrogram

Examples

g <- make_full_graph(5) %du% make_full_graph(5) %du% make_full_graph(5)
g <- add_edges(g, c(1,6, 1,11, 6, 11))
lec <- cluster_leading_eigen(g)
lec

cluster_leading_eigen(g, start=membership(lec))

---