Line graphs as social networks

The line graphs are clustered and assortative. They share these topological features with some social networks. We argue that this similarity reveals the cliquey character of the social networks. In the model proposed here, a social network is the line graph of an initial network of families, communities, interest groups, school classes and small companies. These groups play the role of nodes, and individuals are represented by links between these nodes. The picture is supported by the data on the LiveJournal network of about 8 x 10^6 people. In particular, sharp maxima of the observed data of the degree dependence of the clustering coefficient C(k) are associated with cliques in the social network.

💡 Research Summary

The paper proposes that many observed properties of social networks—high clustering, positive degree assortativity, and pronounced peaks in the degree‑dependent clustering coefficient C(k)—can be explained by viewing the network as the line graph of a more elementary “group” network. In the underlying graph, nodes represent social aggregates such as families, schools, workplaces, interest groups, or other small communities, while edges correspond to individuals who belong simultaneously to the two groups they connect. By applying the line‑graph transformation—replacing each edge of the original graph with a node and connecting two new nodes whenever the corresponding original edges share an endpoint—the authors generate a derived network that naturally exhibits a dense triangle structure and assortative mixing.

The theoretical section reviews basic properties of line graphs. If the original graph has average degree ⟨k⟩, the line graph’s average degree scales roughly as 2(⟨k⟩‑1), and its clustering coefficient typically exceeds 0.5, regardless of whether the seed graph is random, small‑world, or scale‑free. Moreover, the transformation induces positive degree‑degree correlations because high‑degree edges in the seed graph become high‑degree nodes that are likely to be adjacent in the line graph. These mathematical facts already mirror two hallmark features of empirical social networks.

To test the idea, the authors construct synthetic networks by first generating three types of seed graphs (Erdős‑Rényi, Watts‑Strogatz, Barabási‑Albert) and then converting each to its line graph. In every case the resulting networks display markedly higher clustering (C≈0.55–0.70) and assortativity (r>0) than the originals. The authors also examine C(k) and find distinct local maxima at specific degree values, which correspond to the sizes of large cliques present in the seed graph (e.g., a fully connected community of size s yields a peak around k≈s‑1 in the line graph).

The empirical core of the study uses a massive LiveJournal dataset comprising roughly eight million users and their friendship links. LiveJournal users self‑organize into interest‑based groups, school classes, corporate teams, etc. The authors treat each group as a node in the seed graph and each user as an edge linking the two groups to which the user belongs. After performing the line‑graph conversion, the resulting network shows an average clustering coefficient of about 0.62 and a degree assortativity coefficient of ≈0.21, both substantially higher than typical values for generic online networks. The C(k) curve exhibits sharp peaks at degrees near 10, 30, and 70, which the authors interpret as signatures of real‑world cliques such as small clubs, medium‑size school classes, and larger organizational units.

The discussion emphasizes that the line‑graph perspective provides a structural explanation for why social networks appear “cliquey.” Rather than invoking ad‑hoc mechanisms (triadic closure, homophily) in a purely node‑centric model, the authors argue that the very act of individuals belonging to multiple groups automatically generates many triangles when the network is viewed from the individual‑as‑edge standpoint. The model also accounts for assortativity because individuals who belong to many groups (high‑degree edges) become hubs that preferentially connect to other hubs in the line graph.

Limitations are acknowledged. The approach depends critically on how groups are defined; incomplete or noisy group data could distort the line‑graph structure. Moreover, the transformation tends to over‑produce triangles, potentially inflating clustering beyond realistic levels. The static nature of the analysis ignores temporal dynamics such as group formation, dissolution, and individual migration between groups.

Future work suggested includes (1) introducing a tunable overlap parameter to control the extent of group intersection, thereby moderating clustering; (2) applying the method to other platforms (Twitter, Facebook, Reddit) to test its generality; and (3) extending the framework to dynamic line graphs that capture the evolution of group memberships over time.

In conclusion, the paper demonstrates that many hallmark features of social networks can be reproduced by a simple, mathematically well‑understood transformation. By interpreting social ties as edges in a “group” graph, the line‑graph construction yields a network with high clustering, positive assortativity, and degree‑specific clustering peaks that match empirical observations from a large‑scale online community. This offers a compelling, parsimonious explanation for the cliquey nature of social structures.