Fast community detection by SCORE

Consider a network where the nodes split into $K$ different communities. The community labels for the nodes are unknown and it is of major interest to estimate them (i.e., community detection). Degree Corrected Block Model (DCBM) is a popular network model. How to detect communities with the DCBM is an interesting problem, where the main challenge lies in the degree heterogeneity. We propose a new approach to community detection which we call the Spectral Clustering On Ratios-of-Eigenvectors (SCORE). Compared to classical spectral methods, the main innovation is to use the entry-wise ratios between the first leading eigenvector and each of the other leading eigenvectors for clustering. Let $A$ be the adjacency matrix of the network. We first obtain the $K$ leading eigenvectors of $A$, say, $\hat{\eta}_1,\ldots,\hat{\eta}K$, and let $\hat{R}$ be the $n\times (K-1)$ matrix such that $\hat{R}(i,k)=\hat{\eta}{k+1}(i)/\hat{\eta}1(i)$, $1\leq i\leq n$, $1\leq k\leq K-1$. We then use $\hat{R}$ for clustering by applying the $k$-means method. The central surprise is, the effect of degree heterogeneity is largely ancillary, and can be effectively removed by taking entry-wise ratios between $\hat{\eta}{k+1}$ and $\hat{\eta}_1$, $1\leq k\leq K-1$. The method is successfully applied to the web blogs data and the karate club data, with error rates of $58/1222$ and $1/34$, respectively. These results are more satisfactory than those by the classical spectral methods. Additionally, compared to modularity methods, SCORE is easier to implement, computationally faster, and also has smaller error rates. We develop a theoretic framework where we show that under mild conditions, the SCORE stably yields consistent community detection. In the core of the analysis is the recent development on Random Matrix Theory (RMT), where the matrix-form Bernstein inequality is especially helpful.

💡 Research Summary

The paper tackles the problem of community detection in networks whose nodes belong to an unknown set of K communities, focusing on the Degree‑Corrected Block Model (DCBM). DCBM extends the classic Stochastic Block Model by assigning each node i a degree‑heterogeneity parameter θ_i, which multiplies the block‑probability matrix and produces realistic degree variability. This heterogeneity severely hampers traditional spectral clustering methods because the leading eigenvectors of the adjacency matrix are scaled by θ_i, obscuring the community signal.

To overcome this, the authors propose SCORE (Spectral Clustering On Ratios‑of‑Eigenvectors). The algorithm proceeds as follows: (1) compute the first K eigenvectors (\hat\eta_1,\dots,\hat\eta_K) of the adjacency matrix A; (2) form an n × (K‑1) matrix (\hat R) by taking entry‑wise ratios (\hat R(i,k)=\hat\eta_{k+1}(i)/\hat\eta_1(i)) for k = 1,…,K‑1; (3) treat each row of (\hat R) as a (K‑1)‑dimensional feature vector and apply the standard k‑means algorithm to obtain K clusters. The key insight is that the first eigenvector is approximately proportional to the degree‑heterogeneity vector θ, so dividing the other eigenvectors by it cancels the effect of θ. Consequently, the ratio matrix captures only the latent community structure, making the subsequent k‑means step robust to degree variability.

The theoretical contribution rests on recent advances in Random Matrix Theory. By decomposing A into its expectation Ω (derived from the DCBM) and a noise matrix W = A − Ω, the authors invoke a matrix‑form Bernstein inequality to bound (|W|) with high probability. This yields entry‑wise error bounds for the estimated eigenvectors of order (\tilde O(\sqrt{\log n / n})). Because the ratio operation eliminates the common scaling factor, the error in (\hat R) remains of the same order, allowing the authors to prove both weak consistency (mis‑classification rate → 0) and strong consistency (exact recovery with probability tending to one) under mild sparsity and signal‑to‑noise conditions.

Empirically, SCORE is evaluated on two benchmark networks. On a web‑blog dataset with 1,222 nodes and two ground‑truth communities, SCORE misclassifies only 58 nodes, dramatically outperforming classical spectral clustering (≈150 errors) and modularity maximization (≈80 errors). On Zachary’s Karate Club network (34 nodes, two communities), SCORE makes a single mistake, again beating the alternatives.

From a computational standpoint, the dominant cost is the eigen‑decomposition, which can be performed with power iteration or Lanczos methods in roughly O(n log n) time for sparse graphs. Constructing (\hat R) requires O(nK) operations, and k‑means runs in O(nK t) where t is the number of iterations, typically small. Hence the overall algorithm scales linearly (up to log factors) with the number of nodes and is easily applicable to large‑scale networks.

In summary, SCORE offers a conceptually simple yet theoretically solid solution to community detection under DCBM. By exploiting entry‑wise ratios of eigenvectors, it neutralizes degree heterogeneity, achieves provable consistency, delivers superior empirical accuracy, and remains computationally efficient. The paper thus makes a significant contribution to the toolbox of network scientists and data miners dealing with heterogeneous degree structures.