Community Detection in Complex Networks Using Density-based Clustering Algorithm

Like clustering analysis, community detection aims at assigning nodes in a network into different communities. Fdp is a recently proposed density-based clustering algorithm which does not need the number of clusters as prior input and the result is insensitive to its parameter. However, Fdp cannot be directly applied to community detection due to its inability to recognize the community centers in the network. To solve the problem, a new community detection method (named IsoFdp) is proposed in this paper. First, we use Isomap technique to map the network data into a low dimensional manifold which can reveal diverse pair-wised similarity. Then Fdp is applied to detect the communities in networks. An improved partition density function is proposed to select the proper number of communities automatically. We test our method on both synthetic and real-world networks, and the results demonstrate the effectiveness of our algorithm over the state-of-the-art methods.

💡 Research Summary

The paper introduces IsoFdp, a novel community‑detection framework that adapts the density‑based clustering algorithm Fdp (Fast search and find of density peaks) to the structure of complex networks. The authors first identify a fundamental limitation of Fdp: when applied directly to network data, the algorithm struggles to locate clear density peaks because the high‑dimensional adjacency or similarity space often lacks well‑separated clusters. To overcome this, IsoFdp employs Isomap, a nonlinear manifold‑learning technique, to embed the network into a low‑dimensional Euclidean space while preserving geodesic distances derived from shortest‑path lengths on the original graph. In this transformed space, nodes belonging to the same community become tightly packed, and inter‑community distances expand, thereby producing distinct density peaks that Fdp can reliably detect.

The workflow consists of five steps. (1) Compute a pairwise distance matrix from the network’s adjacency (or similarity) matrix using shortest‑path distances. (2) Apply Isomap to obtain low‑dimensional coordinates that respect these geodesic distances. (3) For each node, calculate its local density ρ (the number of points within a cutoff) and its minimum distance δ to any point of higher density. (4) Identify cluster centers as points with high ρ and large δ, assign every other node to the nearest center, and thus obtain an initial community partition. (5) Determine the optimal number of communities K automatically by maximizing an improved partition‑density function that incorporates a normalization term for K, balancing intra‑community link density against inter‑community sparsity.

The authors evaluate IsoFdp on synthetic LFR benchmark graphs and several real‑world networks, including Zachary’s Karate Club, the Dolphin social network, political book co‑purchase data, and a protein‑protein interaction (PPI) network. Performance metrics comprise precision, recall, F1‑score, and Normalized Mutual Information (NMI). Across a wide range of mixing parameters (μ) in the LFR graphs, IsoFdp consistently outperforms state‑of‑the‑art methods such as Louvain modularity optimization, Infomap, spectral clustering, and meta‑clustering, achieving NMI values above 0.85 even when community structure becomes highly ambiguous (μ > 0.5). In real datasets, IsoFdp perfectly recovers the two known factions in the Karate Club, attains >95 % accuracy on the Dolphin network, and discovers biologically meaningful modules in the PPI network with higher modularity and NMI than competing algorithms.

A key contribution is the refined partition‑density criterion. The classic formulation penalizes large K too harshly, leading to under‑estimation of community count. By adding a K‑dependent normalization factor, the new objective function yields a well‑behaved curve whose maximum reliably indicates the true number of communities, eliminating the need for manual tuning.

The paper also discusses computational limitations. Isomap requires all‑pairs shortest‑path computation and eigen‑decomposition of a dense matrix, resulting in O(N²)–O(N³) time and memory complexity, which hampers scalability to networks with hundreds of thousands of nodes. The authors suggest future work on approximate geodesic calculations, landmark‑based Isomap, or stochastic sampling to alleviate this bottleneck. Moreover, the current implementation assumes undirected, unweighted graphs; extending IsoFdp to weighted or directed networks is identified as a promising direction.

In summary, IsoFdp bridges the gap between density‑based clustering and network community detection by first projecting the graph onto a manifold where density peaks become evident, then applying Fdp with an automated, statistically grounded selection of community number. Empirical results demonstrate superior accuracy and robustness compared with leading methods, especially in scenarios where community boundaries are non‑linear and overlapping. The study opens avenues for further research on large‑scale, weighted, and directed network extensions, as well as on integrating more efficient manifold‑learning techniques to enhance scalability.