communities: Community Detection
In cograph: Analysis and Visualization of Complex Networks

communities

R Documentation

Community Detection

Description

Detects communities/clusters in networks using various algorithms. Provides a unified interface to igraph's community detection functions.

Usage

communities(
  x,
  method = c("louvain", "leiden", "fast_greedy", "walktrap", "infomap",
    "label_propagation", "edge_betweenness", "leading_eigenvector", "spinglass",
    "optimal", "fluid"),
  community = NULL,
  weights = NULL,
  resolution = 1,
  directed = NULL,
  seed = NULL,
  ...
)

Arguments

`x`	Network input: matrix, igraph, network, CographNetwork, cograph_network, or tna object
`method`	Community detection algorithm. One of: `"louvain"` - Louvain modularity optimization (default, fast) `"leiden"` - Leiden algorithm (improved Louvain) `"fast_greedy"` - Fast greedy modularity optimization `"walktrap"` - Random walk-based detection `"infomap"` - Information theoretic approach `"label_propagation"` - Label propagation (very fast) `"edge_betweenness"` - Girvan-Newman algorithm `"leading_eigenvector"` - Leading eigenvector method `"spinglass"` - Spinglass simulation `"optimal"` - Exact modularity optimization (slow) `"fluid"` - Fluid communities algorithm
`community`	Optional integer or character vector. If supplied, the returned data frame is filtered to rows whose `community` column matches one of the given values. Default `NULL` (keep all communities).
`weights`	Edge weights. If NULL, uses edge weights from the network if available, otherwise unweighted. Set to NA for explicitly unweighted.
`resolution`	Resolution parameter for modularity-based methods (louvain, leiden). Higher values yield more communities. Default 1.
`directed`	Logical; whether edge-betweenness should treat the network as directed. Default NULL (auto-detect for edge-betweenness). Other methods use their own directed/undirected handling.
`seed`	Random seed for reproducibility. Only applies to stochastic algorithms (louvain, leiden, infomap, label_propagation, spinglass).
`...`	Additional parameters passed to the specific algorithm. See individual functions for details.

Details

When called through this wrapper, methods that require undirected graphs ("louvain", "leiden", "fast_greedy", "leading_eigenvector", and "fluid") fall back to "walktrap" if the input graph is directed.

Algorithm Selection Guide:

Algorithm	Best For	Time Complexity
louvain	Large networks, general use	O(n log n)
leiden	Large networks, better quality than louvain	O(n log n)
fast_greedy	Medium networks	O(n² log n)
walktrap	Networks with clear community structure	O(n² log n)
infomap	Directed networks, flow-based	O(E)
label_propagation	Very large networks, speed critical	O(E)
edge_betweenness	Small networks, hierarchical	O(E² n)
leading_eigenvector	Networks with dominant structure	O(n²)
spinglass	Small networks, allows negative weights	O(n³)
optimal	Tiny networks only (<50 nodes)	NP-hard
fluid	When k is known	O(E k)

Value

A tidy cograph_communities data frame with columns:

node: Node label (character)
community: Community assignment (integer)

Metadata stored as attributes: "algorithm", "modularity", "network" (original input), "igraph_result".

Examples

# Create a network with community structure
if (requireNamespace("igraph", quietly = TRUE)) {
  g <- igraph::make_graph("Zachary")

  # Default (Louvain)
  comm <- cograph::communities(g)
  print(comm)

  # Walktrap
  comm2 <- cograph::communities(g, method = "walktrap")
  print(comm2)
}

cograph documentation built on May 31, 2026, 5:06 p.m.