Finding the proper node ranking method for complex networks

Ranking node importance is crucial in understanding network structure and function on complex networks. Degree, h-index and coreness are widely used, but which one is more proper to a network associated with a dynamical process, e.g. SIR spreading process, is still unclear. To fill this gap, a, which is extracted from the fitting function (f(x)=1-1/(e^(2a(x-b))+1)) of the average number of nodes in each radius of the neighborhood of a node, is proposed. Experiment results which are carried out on twenty real-world networks show that a can classify which of the three measures (degree, h-index and coreness) is more proper to a network in ranking node importance. We also find that [b/3] is a good indicator for forecasting the optimal radius of the neighborhood of a node in ranking node importance for a given network. To the best of our knowledge, it is the first solution of this interesting and open issue. Furthermore, by extending the range of neighborhood where we construct an operator H on of a node, we propose a new method to quantify the importance of a node. The ranking accuracies of most networks can be improved when the radius is increased from 0 to its forecasting optimal radius and the improvement, for the best case, reaches up to 111%. The performances will reduce on half of the networks studied in this paper if we roughly extend the radius of the neighborhood. Our work deepens the understanding of how to find out the proper node ranking method for complex networks. The proposed methods bridge the gaps among network structure and node importance, and may have potential applications in controlling the outbreak of disease, designing of optimal information spreading strategies.

💡 Research Summary

**
The paper addresses a long‑standing open problem in complex‑network analysis: given a network and a dynamical process (here the susceptible‑infected‑recovered, SIR, epidemic model), which of the three widely used node‑importance metrics—degree, h‑index, or coreness—provides the most accurate ranking of influential spreaders? The authors propose a novel, data‑driven approach based on the distribution of the average number of nodes at each distance (radius) from a node, termed ANRN (Average Number of Nodes in each Radius of the Neighborhood).

First, for every network they compute ANRN for all nodes and fit the empirical curve with a logistic‑type function
(f(x)=1-\frac{1}{e^{2a(x-b)}+1}).
The two fitting parameters, a (the steepness) and b (the inflection point), capture essential structural information. Parameter a is shown to be proportional to the maximum variation rate of the fitted curve. By examining twenty real‑world networks from diverse domains (air transportation, power grids, Internet topologies, neural, email, terrorist, social, and collaboration networks), the authors discover that the value of a falls into three distinct intervals. Each interval corresponds to a network class in which one of the three centralities is consistently the best predictor of SIR spreading influence:

• DAN (Degree‑Advantage Network) – degree yields the highest Kendall τ correlation.
• HAN (H‑index‑Advantage Network) – h‑index is superior.
• CAN (Coreness‑Advantage Network) – coreness performs best.

Traditional structural descriptors—average degree, maximum degree, maximum k‑shell index, assortativity, clustering coefficient, density, and graph diameter—do not separate the networks into these three groups, highlighting the novelty of the a‑based classification.

Second, the authors use the second fitting parameter b to forecast the optimal neighborhood radius for ranking. They define the optimal radius as (r_{\text{opt}}=\lfloor b/3\rfloor). To exploit information beyond the immediate neighbors, they introduce a generalized operator H that aggregates the benchmark centrality (degree, h‑index, or coreness) over the 0‑ to (r)-step neighborhood of a node. When (r=0) the operator reduces to the original metric; as (r) grows, the operator gradually incorporates more global structure, analogous to higher‑order h‑indices.

The authors evaluate the performance of the original three metrics and the H‑based extensions across the same twenty networks. For each node they run 200 independent SIR simulations (single seed) and record the average final number of infected/recovered nodes as the ground‑truth influence vector. Kendall τ between a ranking vector and this influence vector quantifies prediction accuracy.

Key findings:

1. For most networks, τ increases monotonically as the radius (r) is enlarged from 0 up to the forecasted (r_{\text{opt}}). The improvement can be dramatic; the best case shows a 111 % increase in τ compared with the baseline (radius 0).
2. Beyond the optimal radius, performance often deteriorates. When the radius is set arbitrarily large (e.g., (r=4) or more), half of the networks experience a drop in τ, indicating that excessive neighborhood aggregation dilutes the local structural signals that are most relevant for epidemic spreading.
3. The optimal radius varies across networks but is typically small (often 2 or 3 steps). This aligns with prior observations that 2‑step neighborhoods capture most of the useful spreading information while keeping computational cost manageable.

The study thus delivers a two‑parameter framework that (i) automatically classifies a network into the most suitable centrality class via a, and (ii) predicts the precise neighborhood size needed to maximize ranking accuracy via b/3. The generalized H operator, applied with the forecasted radius, consistently outperforms the traditional degree, h‑index, and coreness measures.

Implications are broad. In epidemiology, public‑health officials can quickly identify the most effective set of nodes to vaccinate or monitor without exhaustive simulations. In marketing, firms can target the smallest set of influential users for maximal product diffusion. Importantly, the method requires only the network’s adjacency information and the two scalar parameters a and b, making it computationally lightweight even for large, partially observed networks where full global information is unavailable.

Future work could extend the approach to other dynamical processes (e.g., SIS, rumor spreading), weighted or temporal networks, and to directed graphs where in‑ and out‑degrees differ. Nonetheless, the current contribution provides a clear, empirically validated pathway from structural measurement to optimal influence ranking, bridging a gap that has persisted in network science for years.