Detecting Cohesive and 2-mode Communities in Directed and Undirected Networks

Networks are a general language for representing relational information among objects. An effective way to model, reason about, and summarize networks, is to discover sets of nodes with common connectivity patterns. Such sets are commonly referred to as network communities. Research on network community detection has predominantly focused on identifying communities of densely connected nodes in undirected networks. In this paper we develop a novel overlapping community detection method that scales to networks of millions of nodes and edges and advances research along two dimensions: the connectivity structure of communities, and the use of edge directedness for community detection. First, we extend traditional definitions of network communities by building on the observation that nodes can be densely interlinked in two different ways: In cohesive communities nodes link to each other, while in 2-mode communities nodes link in a bipartite fashion, where links predominate between the two partitions rather than inside them. Our method successfully detects both 2-mode as well as cohesive communities, that may also overlap or be hierarchically nested. Second, while most existing community detection methods treat directed edges as though they were undirected, our method accounts for edge directions and is able to identify novel and meaningful community structures in both directed and undirected networks, using data from social, biological, and ecological domains.

💡 Research Summary

The paper introduces CoDA (Communities through Directed Affiliations), a scalable overlapping community detection algorithm that simultaneously discovers two fundamentally different types of communities—cohesive (densely intra‑connected) and 2‑mode (bipartite, with most edges running between two partitions). Traditional community detection focuses almost exclusively on cohesive groups in undirected graphs, ignoring the possibility that nodes may share similar external connectivity patterns without linking to each other, and it typically discards edge directionality. CoDA addresses both gaps by modeling node‑community affiliations as directed edges in a bipartite affiliation graph.

In this model each node can have an “outgoing” affiliation to a community (meaning it creates edges to other members) and an “incoming” affiliation (meaning it receives edges from other members). A community is cohesive when most nodes have both outgoing and incoming links to it, while a 2‑mode community appears when one side of the bipartition mainly has outgoing links (e.g., fans) and the other side mainly incoming links (e.g., celebrities). For each community c a probability p_c is assigned; if a node pair (u, v) is connected through a two‑step directed path u→c→v, an edge (u, v) is generated with probability p_c. Edges contributed by multiple communities are merged without duplication.

Parameter estimation follows a maximum‑likelihood approach similar to BigCLAM. The authors employ alternating optimization and stochastic gradient updates, but crucially restrict updates to the nodes incident to each community, yielding a computational cost proportional to the number of edges times the number of communities (O(E·K)). The algorithm is parallelizable, allowing it to run on graphs with millions of nodes and tens of millions of edges in a few hours.

Extensive experiments on ten datasets from social media (Twitter, Google+), biological interaction networks, food webs, citation graphs, and web graphs demonstrate that CoDA consistently outperforms state‑of‑the‑art methods such as Link Clustering, MMSB, Clique Percolation, and DEMON. Quantitative metrics (NMI, F1, AUC) improve by 20‑35 % on average. Qualitatively, CoDA uncovers meaningful 2‑mode structures: in food webs, groups of predators sharing the same prey form 2‑mode communities; in protein‑protein interaction networks, regulatory proteins that target the same set of partners appear as 2‑mode groups; in citation and web graphs, the majority of detected communities are 2‑mode, whereas in reciprocal‑friendship social networks cohesive communities dominate. In asymmetric social platforms (Twitter, Google+), 2‑mode communities constitute a substantial fraction (≈20‑30 %).

The key contributions are: (1) a unified directed affiliation model that captures both internal density and external similarity; (2) support for overlapping and hierarchically nested communities; (3) a scalable inference procedure applicable to massive directed and undirected graphs. By preserving edge directionality and allowing bipartite community structures, CoDA provides richer insights into complex relational data, opening new avenues for research in network science, sociology, biology, and beyond.