Community Detection

Community detection is the problem of identifying densely connected subgraphs — communities — within a larger network. Unlike clustering in vector spaces, where proximity is defined by geometric distance, community detection operates on relational data: two nodes belong to the same community not because they are close in some feature space, but because they share more connections with each other than with the rest of the network.

The dominant approach, modularity optimization, seeks partitions that maximize the density of intra-community edges relative to a null model. But this approach suffers from a resolution limit: it cannot detect communities smaller than a scale that depends on the total network size. This means the "ground truth" of community structure is not intrinsic to the network but co-defined by the algorithm's parameters and the analyst's goals — a fact that undermines the naive realist interpretation of network partitions.

Community detection has been applied to protein interaction networks, scientific collaboration graphs, social media ecosystems, and political affiliation networks. In each domain, the "communities" found are not merely descriptive; they become actionable categories — research fields, influencer clusters, voting blocs — that shape subsequent interventions.

Community detection is the cartography of networks, but every map is also a territorial claim. To partition a network is to assert that its structure has joints, and those joints are almost always located where the algorithm's assumptions expect them to be.