Shaping Communities out of Triangles

Community detection has arisen as one of the most relevant topics in the field of graph data mining due to its importance in many fields such as biology, social networks or network traffic analysis. The metrics proposed to shape communities are generic and follow two approaches: maximizing the internal density of such communities or reducing the connectivity of the internal vertices with those outside the community. However, these metrics take the edges as a set and do not consider the internal layout of the edges in the community. We define a set of properties oriented to social networks that ensure that communities are cohesive, structured and well defined. Then, we propose the Weighted Community Clustering (WCC), which is a community metric based on triangles. We proof that analyzing communities by triangles gives communities that fulfill the listed set of properties, in contrast to previous metrics. Finally, we experimentally show that WCC correctly captures the concept of community in social networks using real and syntethic datasets, and compare statistically some of the most relevant community detection algorithms in the state of the art.

💡 Research Summary

The paper tackles a fundamental shortcoming in contemporary community‑detection research: most quality functions treat edges merely as a set and ignore how those edges are arranged inside a candidate community. In social networks, the presence of closed triangles (three mutually connected vertices) is widely recognized as a proxy for trust, shared interests, and cohesive interaction. The authors therefore argue that any meaningful definition of a community should explicitly account for the density and distribution of triangles within it.

First, the authors formalize three desiderata for a “good” community in social graphs: (1) Cohesiveness – a community should contain many internal triangles; (2) Structure – triangles should be evenly distributed so that the community is not just a collection of dense sub‑clusters but a well‑structured whole; (3) Well‑definedness – the community’s boundary should be sharp, meaning that vertices have few triangle‑based connections to the outside. These properties are deliberately chosen to capture aspects that traditional metrics such as modularity, conductance, or cut‑ratio cannot express.

To satisfy these criteria, the authors introduce Weighted Community Clustering (WCC). For a community (C) and a vertex (v\in C), let (t_{\text{in}}(v,C)) be the number of triangles that lie completely inside (C) and involve (v), and let (t_{\text{total}}(v)) be the total number of triangles that involve (v) in the whole graph. The community score is defined as

\