Pith. sign in

REVIEW 4 major objections 4 minor 28 references

HGC: A hybrid method combining gravity model and cycle structure for identifying influential spreaders in complex networks

T0 review · 4 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper proposes HGC, a centrality score that combines a gravity model with effective distance and cycle structure, and reports that it identifies influential spreaders more accurately than seven comparison methods on eight real-world…

desk verdict A coherent hybrid centrality that is competitive with, not clearly better than, LGM; the abstract overstates what Table 2 shows. read the letter →

arxiv 2411.17138 v1 pith:6HYCLJFF submitted 2024-11-26 cs.CE

Jiaxun Li , Yonghou He , Zhefan Dong , Li Tao This is my paper · ORCID
classification cs.CE
keywords complexnetworksinfluentialspreadersgravitymodeleffectivedistancecyclestructureSIRcentralityrankinghybridmethod
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to establish that a node's capacity to spread influence is better captured by combining the gravity model's mass-and-distance logic with an asymmetric distance and with cycle-structure information. To do this it proposes the HGC score, which sums a gravity interaction term $GM(i)$ and a cycle-ratio propagation term $RCP(i)$, scaled by a balancing factor $\gamma$. On eight real-world networks, with SIR simulations as the ground truth for spreading, HGC achieves higher Kendall's tau values than degree, betweenness, closeness, k-shell, cycle ratio, local gravity, and RDP, and it also improves top-k Jaccard similarity, the spreading power of its top nodes, and ranking monotonicity. The paper concludes that replacing topological shortest-path distance with effective distance fixes the symmetry flaw of prior gravity models, and that cycle structure adds redundant-pathway information that matters in densely connected networks.

What carries the argument

The machinery is a two-term centrality. First, a gravity term $GM(i)=\sum_{ED(j|i)\le R} e^{-c(i)} k_i k_j / ED_{j|i}^2$, where $c(i)$ is Burt's network constraint coefficient for structural holes, $k$ is degree, and effective distance $ED_{j|i}=1-\log_2(a_{ij}/k_i)$ replaces shortest-path distance. Second, a propagation term $RCP(i)=\sum_{l=1}^{T} I_l(i)$ with $I_l(i)=l^{-2}\sum_{j\in\Lambda_i} I_{l-1}(j)$, seeded by the cycle ratio $CR(i)$, which quantifies involvement in other nodes' shortest cycles. The balancing factor $\gamma=\langle GM\rangle/\langle RCP\rangle$ fuses the two scores. Taken together, the machinery encodes three ideas: influence decays with probability-based asymmetric distance, bridging positions (structural holes) add weight, and nodes embedded in cycles inherit propagation power from their neighbours.

What would settle it

Take a tree network (no cycles at all): HGC's RCP term is zero for every node, so HGC collapses to its gravity term; if HGC then still outperforms LGM by the same margin it shows on cyclic networks, the cycle component is not the source of the claimed advantage. Alternatively, rerun the Facebook comparison with a different shortest-cycle enumeration and check whether the top-k list changes materially.

Watch

Extended reading notes

Core claim

The central discovery is a single score, $HGC(i) = GM(i) + \gamma RCP(i)$, where $GM(i)$ accumulates $e^{-c(i)} k_i k_j / ED_{j|i}^2$ over neighbours within a radius $R$, $RCP(i)$ spreads each node's cycle ratio through $l^2$-damped neighbour iterations, and $\gamma = \langle GM\rangle/\langle RCP\rangle$ balances the two scales. The effective distance $ED_{j|i}=1-\log_2 P_{j|i}$, with $P_{j|i}=a_{ij}/k_i$, makes the distance from $i$ to $j$ different from the distance from $j$ to $i$, so influence between a pair is no longer symmetric. The cycle ratio $CR(i)$ counts how strongly node $i$ participates in the shortest cycles of neighbouring nodes, and the RCP iteration carries that structural information to higher-order neighbours. The paper argues that these two ingredients, asymmetric effective distance and cycle-based propagation, are what let HGC identify nodes that SIR spreading actually starts from better than the seven baselines.

Load-bearing premise

The method assumes that the number of shortest cycles passing through each pair of nodes can be computed for any network, but no algorithm or complexity bound is given for that computation, so the method's rankings on large networks like Facebook and Power depend on an unspecified cycle-enumeration choice.

Editorial extensions

If this is right

  • HGC gives more accurate full-network rankings than the seven baselines on five of eight test networks, with second-best performance on the sparse Power grid.
  • The top-10 nodes chosen by HGC produce faster-growing SIR infection curves than baseline top-10 sets in Jazz, USAir, NS, and Facebook.
  • HGC's monotonicity is the highest of all methods on every network studied, so its scores separate nodes more finely and produce fewer tied ranks.
  • Because the method uses only second-order neighbourhoods ($R=2$) and two propagation iterations ($T=2$), it remains a local computation once cycle ratios are known.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper gives no algorithm or complexity bound for computing the cycle-number matrix $c_{ij}$, so on large networks such as Facebook the reported rankings depend on an unspecified shortest-cycle enumeration; this leaves the method's practical scalability open.
  • The asymmetric effective distance could be carried directly into directed or weighted networks, where influence asymmetry is the norm, though the paper's experiments are limited to undirected unweighted graphs.
  • A natural ablation would compare HGC against its own gravity-only term: if most of the Kendall-tau gain survives without $RCP(i)$, then cycle structure is not the decisive ingredient the narrative suggests.
  • The balancing factor $\gamma$ is data-dependent, and on near-tree networks where cycle ratios vanish the RCP term collapses, so behaviour on very sparse networks is an untested regime.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes HGC, a hybrid centrality for identifying influential spreaders in undirected unweighted networks. HGC combines a gravity-model interaction term using effective distance and structural-hole constraint with an information-propagation term based on cycle ratio, fused through a balancing factor. The method is evaluated on eight real-world networks using SIR simulations and compared with seven baselines in terms of Kendall's tau, Jaccard similarity, and monotonicity. The authors claim that HGC outperforms all seven compared methods.

Significance. If the claims were fully supported, HGC would be a useful contribution to influential-spreader identification, since it combines asymmetric effective-distance information with cycle-structure information rather than fitting free parameters to SIR outcomes. The paper has strengths: evaluation on multiple real networks, comparison with several established baselines, and no circular fitting of the target result. However, the headline claim of universal superiority is contradicted by the paper's own table, and the method's core interaction term appears to be disabled by the chosen truncation radius as written. These issues must be resolved before the contribution can be assessed.

major comments (4)
  1. [Abstract; Section 5.1, Table 2] The opening claim that HGC outperforms all seven compared methods is not supported by Table 2. LGM attains higher Kendall's tau on USAir (0.7758 vs 0.766), Yeast (0.8099 vs 0.8085), and Power (0.7383 vs 0.7145), and on the five networks where HGC leads, the advantages over LGM are between 0.005 and 0.018. The SIR reference values are stochastic averages of 1000 runs, yet no confidence intervals, standard deviations, or significance tests are reported. The abstract and conclusion should either be narrowed to "competitive with" or "best on five of eight," or the margins need statistical support.
  2. [Section 3.1.1, Eq. (5)] The truncation radius R=2 appears incompatible with the effective-distance definition. From Eq. (1), the effective distance from a source i of degree k_i to any adjacent node j is 1 - log2(1/k_i) = 1 + log2(k_i), which exceeds 2 whenever k_i > 2. Therefore, under the condition ED(j|i) ≤ R in Eq. (5), GM(i)=0 for all nodes of degree greater than 2, and the gravity/interaction component contributes nothing for typical nodes in the datasets (e.g., USAir has average degree 12.8). Please clarify whether R is a hop count or an effective-distance threshold; if the latter, the experiments need to be redone with a meaningful R, and if the former, Eq. (5) must be revised.
  3. [Section 3.1.2, Eq. (8), Algorithm 1] The computation of the cycle number matrix c_ij is not specified. The paper only states that c_ij counts shortest cycles through i and j, but gives no algorithm, data structure, or complexity bound. On Facebook (4039 nodes) and Power (4941 nodes), enumerating all shortest cycles through every pair is nontrivial; if an approximation or a cycle basis is used instead, the reported rankings depend on an unspecified choice. A complete and reproducible specification of this step is required.
  4. [Section 5, Tables 2 and 3] No sensitivity or ablation analysis is provided for the fixed choices R=2, T=2, and the balancing factor gamma. Since gamma is set as a ratio of averages and is not tuned to the SIR outcome, this is not circular, but the narrow margins in Table 2 leave open the possibility that the results depend heavily on these choices. Reporting results for R=1,2,3 and T=1,2,3, together with an ablation that removes the GM or RCP term, would substantiate the claimed contribution of each component.
minor comments (4)
  1. [Section 4.4, Eq. (11)] The Jaccard similarity coefficient is defined for sets, but the text refers to sequences; please clarify that the top-k lists are converted to sets and state the range of k used in Figure 4.
  2. [Section 5.3, Table 3] The text says RDP achieved the best performance in the Email network and that RDP and LGM attained the highest values in PB and Facebook, but Table 3 shows HGC tied for first in all of those networks; please correct the description.
  3. [Figures 4 and 5] The axis labels and legends in Figures 4 and 5 appear garbled in the manuscript version; please provide high-resolution figures with readable text.
  4. [Algorithm 1] Line 12 computes the balancing factor gamma inside the neighbor loop; this should be computed once after the GM and RCP accumulations are complete.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: HGC is an independently evaluated heuristic; no parameter is fitted to the SIR target.

full rationale

The derivation chain of HGC is self-contained and does not reduce to its evaluation target. The method score HGC(i) = GM(i) + γ RCP(i) combines a gravity-model interaction term GM(i) (Eq. 5), built from degree, effective distance, and the network constraint coefficient, with a cycle-ratio propagation term RCP(i) (Eqs. 6–7) built from the cycle ratio CR(i) (Eq. 8). The balancing factor γ = ⟨GM⟩/⟨RCP⟩ (Eq. 9) is computed from the method's own component scores, so it is used only to put the two terms on a common scale; it is not fitted to SIR outcomes or to the evaluation rankings. The truncation radius R = 2 and iteration count T = 2 are fixed in advance with a stated small-world justification, not tuned against the epidemic simulations. The evaluation compares the resulting ranking against independent SIR simulations using Kendall's tau, Jaccard similarity, and monotonicity; no fitted parameter is renamed as a prediction, and no definition of the method presupposes the SIR ranking it aims to predict. The cited prior work (cycle ratio, gravity model, effective distance) supplies building blocks, but the HGC combination is a new heuristic, and the empirical claim of outperformance — even if contestable on statistical grounds, as on USAir, Yeast, and Power — is a claim about benchmark results, not a circular derivation. Therefore no circularity step can be exhibited from the paper's own equations.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The method rests on imported definitions (effective distance, gravity, cycle ratio, structural holes, SIR) rather than new postulates. No free parameter is fitted to the target SIR rankings, but gamma, R, and T are hand-set. The least-supported element is the computability and implementation of the cycle number matrix.

free parameters (3)
  • Balancing factor gamma = network-dependent ratio mean(GM)/mean(RCP), not reported
    Eq. 9 defines gamma as a global scale-matching constant; it is not fitted to SIR labels but is a hand-defined normalization that affects the final score.
  • Truncation radius R = 2
    Algorithm 1 sets R=2 for all networks, chosen a priori due to small-world property; affects which neighbors enter Eq. 5.
  • Iterations T = 2
    Algorithm 1 sets T=2 for the cycle-ratio propagation in Eq. 7; increasing T would change rankings.
assumptions (5)
  • domain assumption Effective distance defined by transition probabilities (ED_{j|i}=1-log2(P_{j|i})) is a valid distance metric for shortest-path aggregation
    Imported from Brockmann and Helbing (ref [25]); the gravity model sums nodes within ED <= R, treating ED as a distance.
  • domain assumption Gravity analogy: influence between nodes scales as product of degrees divided by squared effective distance
    Used in Eq. 5; this is the standard gravity-model assumption from refs [12]-[16].
  • domain assumption Cycle ratio CR(i) defined in Eq. 8 captures node importance via involvement in shortest cycles
    Imported from Fan et al. (ref [8]); the paper uses it without re-derivation.
  • ad hoc to paper The cycle number matrix c_ij is computable for arbitrary large graphs
    The paper provides only a toy example (three shortest cycles) and no algorithm for enumerating shortest cycles in real networks with thousands of nodes.
  • domain assumption SIR model with recovery probability lambda=1 and beta=beta_c is the ground truth for spreading influence
    Section 4.3; all accuracy claims are measured against this simulation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HGC: A hybrid method combining gravity model and cycle structure for identifying influential spreaders in complex networks." pith.science (2026). https://pith.science/paper/6HYCLJFF

@misc{pith2026241117138,
  author       = {Pith},
  title        = {Pith review of: HGC: A hybrid method combining gravity model and cycle structure for identifying influential spreaders in complex networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6HYCLJFF}},
  note         = {Machine review of arXiv:2411.17138}
}
read the original abstract

Identifying influential spreaders in complex networks is a critical challenge in network science, with broad applications in disease control, information dissemination, and influence analysis in social networks. The gravity model, a distinctive approach for identifying influential spreaders, has attracted significant attention due to its ability to integrate node influence and the distance between nodes. However, the law of gravity is symmetric, whereas the influence between different nodes is asymmetric. Existing gravity model-based methods commonly rely on the topological distance as a metric to measure the distance between nodes. Such reliance neglects the strength or frequency of connections between nodes, resulting in symmetric influence values between node pairs, which ultimately leads to an inaccurate assessment of node influence. Moreover, these methods often overlook cycle structures within networks, which provide redundant pathways for nodes and contribute significantly to the overall connectivity and stability of the network. In this paper, we propose a hybrid method called HGC, which integrates the gravity model with effective distance and incorporates cycle structure to address the issues above. Effective distance, derived from probabilities, measures the distance between a source node and others by considering its connectivity, providing a more accurate reflection of actual relationships between nodes. To evaluate the accuracy and effectiveness of the proposed method, we conducted several experiments on eight real-world networks based on the Susceptible-Infected-Recovered model. The results demonstrate that HGC outperforms seven compared methods in accurately identifying influential nodes.

Figures

Figures reproduced from arXiv: 2411.17138 by the authors.

Figure 1
Figure 1. HGC Overall Framework Diagram. The overall framework of HGC comprises three parts: (1) interaction effects between nodes based on the gravity model; (2) information propagation effects based on the cycle structure; and (3) weighted fusion. The framework can compute the influence scores of all nodes in the network through the three parts outlined above.     0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 2 … view at source ↗
Figure 2
Figure 2. Figure. (a) shows the cycle number matrix of the example network in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Schematic diagram of SIR model state transition. 4.4. Evaluation indicators The Kendall’s Tau [22] is a statistical measure used to assess the correlation between two ranked sequences. It reflects the degree of association between two variables by calculating the consistency of their orderings. If in both sequences, the element ranked higher/lower in the first sequence is also ranked higher/lower in the second seque… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: The Jaccard similarity coefficients on the top-k influential spreaders. The X-axis represents the number of top influential spreaders, and the Y-axis represents the value of the Jaccard similarity coefficient [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: The propagation influence of the top-10 ranking nodes of various methods simulated by SIR (under β = βc). The X-axis represents the time step, and the Y-axis represents the number of infected and recovered nodes at time t [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 13 canonical work pages

  1. [1]

    A. I. E. Hosni, K. Li, S. Ahmad, Minimizing rumor influence in multiplex online social networks based on human individual and social behaviors, Information Sciences 512 (2020) 1458–1480, https://doi.org/10.1016/j.ins.2019.10.063

    work page doi:10.1016/j.ins.2019.10.063 2020

  2. [2]

    W. Xu, T. Li, W. Liang, J. X. Yu, N. Yang, S. Gao, Identifying structural hole spanners to maximally block information propagation, Information Sciences 505 (2019) 100–126, https://doi.org/10.1016/J.INS.2019.07.072

    work page doi:10.1016/j.ins.2019.07.072 2019

  3. [3]

    P. Zhu, Q. Zhi, Y . Guo, Z. Wang, Analysis of epidemic spreading process in adaptive networks, IEEE Transactions on Circuits and Systems II: Express Briefs 66 (2018) 1252–1256, https://doi.org/10.1109/TCSII.2018.2877406

    arXiv 2018

  4. [4]

    S. Yao, N. Fan, J. Hu, Modeling the spread of infectious diseases through influence maximization, Optimization Letters 16 (2022) 1563–1586, https://doi.org/10.1007/s11590-022-01853-1

    work page doi:10.1007/s11590-022-01853-1 2022

  5. [5]

    S. S. Chaharborj, K. N. Nabi, K. L. Feng, S. S. Chaharborj, P. S. Phang, Controlling COVID-19 transmission with isolation of influential nodes, Chaos, Solitons & Fractals 159 (2022) 112035, https://doi.org/10.1016/j.chaos.2022.112035

    arXiv 2022

  6. [6]

    L. C. Freeman, A set of measures of centrality based on betweenness, Sociometry (1977) 35–41 https://doi.org/10.2307/3033543

    doi:10.2307/3033543 1977

  7. [7]

    L. C. Freeman, et al., Centrality in social networks: Conceptual clarification, Social Network: Critical Concepts in Sociology. Londres: Routledge 1 (2002) 238–263, https://doi.org/10.1016/0378-8733(78)90021-7

    doi:10.1016/0378-8733(78)90021-7 2002

  8. [8]

    T. Fan, L. L ¨u, D. Shi, T. Zhou, Characterizing cycle structure in complex networks, Communications Physics 4 (2021) 272, https://doi.org/10.1038/s42005-021-00781-3

    work page doi:10.1038/s42005-021-00781-3 2021

Show all 28 references
  1. [9]

    P. Bonacich, Factoring and weighting approaches to status scores and clique identification, Journal of Mathematical Sociology 2 (1972) 113–120, https://doi.org/10.1080/0022250X.1972.9989806

    1972

  2. [10]

    L. L ¨u, T. Zhou, Q.-M. Zhang, H. E. Stanley, The H-index of a network node and its relation to degree and coreness, Nature Communications 7 (2016) 10168, https://doi.org/10.1038/ncomms10168

    2016 doi

  3. [11]

    Z. Li, X. Huang, Identifying influential spreaders using local information, Mathematics 11 (2023) 1302, https://doi.org/10.3390/math11061302

    2023 doi

  4. [12]

    L.-l. Ma, C. Ma, H.-F. Zhang, B.-H. Wang, Identifying influential spreaders in complex networks based on gravity formula, Physica A: Statistical Mechanics and its Applications 451 (2016) 205–212, https://doi.org/10.1016/j.physa.2015.12.162

    2016 doi

  5. [13]

    Z. Li, T. Ren, X. Ma, S. Liu, Y . Zhang, T. Zhou, Identifying influential spreaders by gravity model, Scientific Reports 9 (2019) 8387, https://doi.org/10.1038/s41598-019-44930-9

    2019 doi

  6. [14]

    Z. Li, X. Huang, Identifying influential spreaders in complex networks by an improved gravity model, Scientific Reports 11 (2021) 22194, https://doi.org/10.1038/s41598-021-01218-1

    2021 doi

  7. [15]

    Z. Li, X. Huang, Identifying influential spreaders by gravity model considering multi-characteristics of nodes, Scientific Reports 12 (2022) 9879, https://doi.org/10.1038/s41598-022-14005-3

    2022 doi

  8. [16]

    S. Zhu, J. Zhan, X. Li, Identifying influential nodes in complex networks using a gravity model based on the H-index method, Scientific Reports 13 (2023) 16404, https://doi.org/10.1038/s41598-023-43585-x . 11

    2023 doi

  9. [17]

    J. T. Lizier, F. M. Atay, J. Jost, Information storage, loop motifs, and clustered structure in complex networks, Physical Review E—Statistical, Nonlinear, and Soft Matter Physics 86 (2012) 026110, https://doi.org/10.1103/PhysRevE.86.026110

    2012 doi

  10. [18]

    D. Shi, G. Chen, W. W. K. Thong, X. Yan, Searching for optimal network topology with best possible synchronizability, IEEE Circuits and Systems Magazine 13 (2013) 66–75, https://doi.org/10.1109/MCAS.2012.2237145

    2013

  11. [19]

    Ruths, D

    J. Ruths, D. Ruths, Control profiles of complex networks, Science 343 (2014) 1373–1376, https://doi.org/10.1126/science.1242063

    2014 doi

  12. [20]

    R. S. Burt, Structural holes and good ideas, American Journal of Sociology 110 (2004) 349–399, https://doi.org/10.1086/421787

    2004 doi

  13. [21]

    H. W. Hethcote, The mathematics of infectious diseases, SIAM Review 42 (2000) 599–653, https://doi.org/10.1137/S0036144500371907

    2000 doi

  14. [22]

    M. G. Kendall, A new measure of rank correlation, Biometrika 30 (1938) 81–93, https://doi.org/10.1093/BIOMET/30.1-2.81

    1938 doi

  15. [23]

    Zareie, A

    A. Zareie, A. Sheikhahmadi, M. Jalili, M. S. K. Fasaei, Finding influential nodes in social networks based on neighborhood correlation coefficient, Knowledge-based systems 194 (2020) 105580, https://doi.org/10.1016/j.knosys.2020.105580

    2020

  16. [24]

    J. Bae, S. Kim, Identifying and ranking influential spreaders in complex networks by neighborhood coreness, Physica A: Statistical Mechanics and its Applications 395 (2014) 549–559, https://doi.org/10.1016/j.physa.2013.10.047

    2014 doi

  17. [25]

    Brockmann, D

    D. Brockmann, D. Helbing, The hidden geometry of complex, network-driven contagion phenomena, Science 342 (2013) 1337–1342, https://doi.org/10.1126/science.1245200

    2013 doi

  18. [26]

    D. J. Watts, S. H. Strogatz, Collective dynamics of ‘small-world’ networks, Nature 393 (1998) 440–442, https://doi.org/10.1038/30918

    1998 doi

  19. [27]

    Kitsak, L

    M. Kitsak, L. K. Gallos, S. Havlin, F. Liljeros, L. Muchnik, H. E. Stanley, H. A. Makse, Identification of influential spreaders in complex networks, Nature Physics 6 (2010) 888–893, https://doi.org/10.1038/nphys1746

    2010 doi

  20. [28]

    Castellano, R

    C. Castellano, R. Pastor-Satorras, Thresholds for epidemic spreading in networks, Physical Review Letters 105 (2010) 218701, https://doi.org/10.1103/PhysRevLett.105.218701. 12

    2010 doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.