{"id":"97ea1509-e2e7-4f8f-aaf3-b6a63e7dc4eb","arxiv_id":"1908.04901","paper_version":3,"verdict":"UNVERDICTED","confidence":"HIGH","novelty_score":0.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A position paper summarizing generative models, dynamic community detection, and community-aware immunization strategies, with the takeaway that exploiting community structure improves performance.","lead":"This paper reviews three topics in community structure research: generative models and modularity, time-evolving communities, and immunization strategies for modular networks. It is useful as a survey and research agenda for newcomers and as a reference for practitioners seeking to incorporate community structure into network analysis.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central comparative claim in §4.3.5 rests on cross-paper rankings from heterogeneous SIR/SI experiments; no unified benchmark controls epidemic parameters, network generators, or immunization budgets.","rationale":"The reader identified the same premise as least secure: the comparative conclusions assume that SIR/SI simulations and synthetic benchmarks across the cited papers are mutually consistent and representative of real contact networks. I agree with that assessment. The concern is not that any cited study is wrong, but that the review's central comparative claim overreaches the evidence by converting heterogeneous, uncontrolled experimental comparisons into a general ranking. Because this is a position paper and survey rather than a new research contribution, the appropriate verdict remains UNVERDICTED: the review is useful as a roadmap, but its central synthesis requires a unified controlled benchmark before it can be accepted as a robust finding. I therefore do not change the reader's verdict.","tokens_in":29663,"tokens_out":3189,"duration_ms":35868,"concrete_test":"Run a single controlled benchmark: generate LFR networks with N=5000, average degree 10, mixing parameter mu in {0.1, 0.3, 0.5, 0.7}, use ground-truth communities, simulate SIR with a fixed transmissibility (e.g., R0=1.5) and immunization budget of 5% of nodes, and compare final epidemic size for one representative local strategy (community centrality or Super node), one global strategy (NNC or Bridgeness), one combined strategy (CbM or WCHB), and one stochastic strategy (BHD or RWOS). If the predicted crossing of local versus global performance does not appear, or the combined strategy is not uniformly best, then the Section 4.3.5 synthesis is an artifact of heterogeneous study setups.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The review's central synthesis, stated in Section 4.3.5, is that immunization performance increases with greater use of community-structure information and that local strategies outperform global strategies in well-separated communities while global strategies win in loose communities. The paper itself runs no experiment; these conclusions are aggregated from independent studies that differ in epidemic model (SIR, SI, independent cascade), network generator (LFR, Facebook subnetworks, co-authorship graphs), community-detection method, immunization budget, and baseline strategies. A strategy's reported rank may therefore reflect the experimental setup rather than intrinsic method quality. For example, the claimed superiority of BHD and RWOS over CBF in stochastic strategies, and of combined strategies such as CbM and WCHB in deterministic strategies, is never tested on a common footing. The phrase 'more information about the community structure' is also underspecified: it conflates membership counts, inter-community link proportions, community sizes, and bridge-hub identities, which have no common scale. Consequently, the conditional local/global ranking asserted in Section 4.3.5 is not established by the evidence presented in the review.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript is a review/position paper in three parts. Section 2 reviews generative models for community structure (ER graphs, configuration model, stochastic block model and its degree-corrected variants, planted partition model) and statistical inference, including the connection between modularity maximization and maximum-likelihood estimation of the planted partition model. Section 3 surveys dynamic community detection, organized into snapshot-based, evolutionary, incremental/online, and prediction-oriented approaches. Section 4 reviews immunization strategies for modular networks, classifying them into stochastic and deterministic strategies and, within the latter, into global, local, and combined variants for non-overlapping and overlapping communities. The paper's central synthesis, stated in Section 4.3.5 and echoed in Section 5, is that immunization performance increases as more community-structure information is exploited, that local strategies outperform global strategies in networks with strong community structure, that global strategies outperform local strategies in networks with loose community structure, and that combined strategies generally perform best. The review also identifies open problems, notably the lack of controlled benchmarks for evolving community detection.","tokens_in":29874,"tokens_out":8643,"duration_ms":81493,"significance":"The review is competently assembled and the mathematical core (Eqs. 1-10) is standard and correctly transcribed from the cited literature. It provides a useful taxonomy of the immunization literature and a clear statement of open problems, such as the need for benchmarks for dynamic community detection. The paper's contribution is synthetic rather than novel: it runs no unified experiment, and it is transparent about shipping no code, which is acceptable for a survey. If the Section 4.3.5 synthesis were established on a common benchmark, it would be practically valuable, since it would give practitioners clear guidance on when to prefer community-aware strategies and how much structural information is worth collecting. The main weakness is evidential: the central comparative claims are aggregated from many heterogeneous studies rather than demonstrated on a common footing, so the review's conclusions need to be either strengthened by a systematic comparison or explicitly qualified.","major_comments":[{"comment":"The paper's central claim that 'the performance of the immunization strategies increases when more information about the community structure is used' and the conditional local/global ranking are not established by the evidence presented. The strategies being compared (CBF, DCBF, BHD, RWOS, Mod, BVA, NNC, CbM, WCHB, OC, and others) come from independent studies that differ in epidemic model (SIR vs SI vs independent cascade), network generator (LFR, Facebook subnetworks, co-authorship graphs), community-detection algorithm, immunization budget, and baseline strategy. A reported rank can therefore reflect the experimental setup rather than intrinsic method quality; for example, the claimed superiority of BHD and RWOS over CBF, or of WCHB over Comm and CbM, is never tested on a common footing. To make the synthesis load-bearing, the authors should either (i) tabulate for every cited comparison the epidemic model, network type, budget, and baseline and restrict each ranking claim to matched settings, or (ii) explicitly soften the claims to 'within each cited study, more community information helped.' Without this, Section 4.3.5 overstates what the literature supports.","section":"§4.3.5 and §4.1.2"},{"comment":"The conditional ranking depends on the notion of 'community structure strength,' which is used informally and inconsistently across the review: in §4.1.2 it is identified with high modularity (Q > 0.84), in §4.3.1 with the proportion of intra-community links, and in §4.3.3 with 'medium strength' without any formal threshold. Since the central synthesis says that strategy choice should depend on this quantity, the review needs a working definition, or at least a statement that the cited studies measure it in incompatible ways. As written, the conditional local/global ranking is not falsifiable from the survey data.","section":"§4.3.5"},{"comment":"The phrase 'more information about the community structure' is used in incompatible senses: membership counts (RWOS), inter-community link proportions (WCHB), community sizes (CbC), and bridge-hub identities (BHD) are treated as if they lay on a single scale of information content. The claim that performance increases with more information therefore conflates qualitatively different features. The authors should either define a partial order over the information features used by the discussed strategies or restrict the statement to specific features, otherwise the central synthesis is unfalsifiable.","section":"§4.3.5"}],"minor_comments":[{"comment":"Figure 1 cites CBF [3], DCBF [7], BHD [4], and RWOS [8], but the text cites these methods as [11], [89], [12], and [90]; Figure 2 has analogous mismatches (for example, 'Community centrality [2]' whereas the text discusses it as [10]). Please update all figure citations to the manuscript's reference list.","section":"Figures 1 and 2"},{"comment":"Section 5 contains the sentence 'Another drawback of this approach is that the stochastic block model requires the selection of the number of communities...' twice in consecutive paragraphs; please delete the duplicate.","section":"§5"},{"comment":"The acronym 'WCBM' appears in this section, while the same strategy is defined earlier as 'WCHB' (also written 'WCBH' in places); please unify the acronym throughout.","section":"§4.3.5"},{"comment":"The text says the modular centrality work 'has been extended to networks with non-overlapping community structure [109]', but reference [109] is titled 'Centrality in complex networks with overlapping community structure'; please correct the wording.","section":"§4.3.4"},{"comment":"The Metropolis-Hastings acceptance probability is written as a = min{...} without the upper bound of 1; as written, a can exceed 1. The expression should be min(1, ...).","section":"Eq. (9)"},{"comment":"The notation '{i,j}∈r' is ambiguous regarding whether ordered or unordered pairs are summed; because the factor of two matters in the modularity expression, please specify this explicitly.","section":"Eqs. (2) and (8)"},{"comment":"The label 'K-sell with community' should read 'k-shell with community'.","section":"Figure 2"}],"recommendation":"major_revision","confidential_remarks":"The comparative section leans heavily on the authors' own prior papers (notably Ghalmane, Cherifi, and coauthors). This is not inappropriate for a survey, but the editor may wish to seek independent confirmation of the ranking claims, since no unified benchmark is provided. The manuscript is best characterized as an expository survey/position paper; its fit for the journal depends on whether such synthesis pieces are within scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a position paper and survey, not a research preprint. No new equations, algorithms, or data. Its value is organizational: it puts generative models, temporal community detection, and immunization strategies into a usable taxonomy, and the Section 4 classification (stochastic vs deterministic, local/global/combined, overlapping vs non-overlapping) is genuinely handy. The math in Section 2 is standard and correctly transcribed; the SBM/planted partition/modularity equivalence discussion is accurate. The temporal section is a fair, compact summary of snapshot, evolutionary, incremental, and predictive approaches.\n\nThe soft spot is the one the stress-test flags. The central comparative claim in Section 4.3.5—that immunization performance increases with more community-structure information, and that local strategies beat global in strong-community networks while global beat local in loose ones—is presented as a synthesis of experiments that were never run on a common footing. The review aggregates SIR/SI simulations across different generators (LFR, Facebook subnetworks, co-authorship graphs), different epidemic parameters, different community-detection methods, and different immunization budgets. That makes the cross-paper ranking of methods like BHD vs CBF vs RWOS, or CbM vs WCHB, unreliable as evidence. The phrase “more information about the community structure” is also slippery: it covers membership counts, bridge counts, community sizes, and inter-community link fractions, which don't share a scale. So the conditional ranking is plausible but not established by the evidence this paper presents.\n\nThat said, the paper itself mostly knows its status. It calls itself a position paper; it doesn't claim to run a benchmark. The appropriate fix is modest: soften the summary claims in §4.3.5 and the conclusion, flag the heterogeneity of the underlying experiments, and maybe add a short paragraph on what a unified benchmark would need to control. There are also small editorial issues—a duplicated sentence in Section 5 and the “WCBM” typo—but nothing load-bearing.\n\nWho is this for: grad students or researchers wanting a map of the community-detection and immunization literature. It deserves a serious referee, because a good survey with a strong taxonomy is worth publishing, but I would not treat its comparative performance claims as established results. If I cite it, it will be for the taxonomy, not for the ranking.","headline":"A useful survey and taxonomy of community detection and immunization strategies, but its central comparative claim about immunization performance rests on heterogeneous experiments never run on a common footing.","tokens_in":30393,"tokens_out":1776,"would_cite":true,"duration_ms":16913,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Community structure should shape epidemic immunization strategy","keywords":["community structure","modularity","stochastic block model","community detection","time-evolving networks","immunization strategies","epidemic spreading","centrality"],"falsifier":"Run a single controlled benchmark on networks that differ only in community strength, holding degree sequence, size, and epidemic parameters fixed, and compare one local strategy with one global strategy. If the global strategy wins in strongly modular networks, or the local strategy wins in loosely modular networks, the paper's central conditional ranking is falsified; observing no monotone relationship between added community information and epidemic size would also undercut it.","tokens_in":29502,"feed_emoji":"💉","tokens_out":4829,"duration_ms":48678,"temperature":0.7,"pith_summary":"This position paper argues that community structure is not a byproduct of network analysis but a load-bearing feature that should drive how we model networks, track their evolution, and control epidemics. The paper synthesizes generative models, chiefly the stochastic block model and its degree-corrected variant, and shows how they connect to modularity maximization, including a recent equivalence result and bounds on the resolution parameter. For time-evolving networks, it catalogs snapshot matching, evolutionary algorithms, and incremental and online methods. Its central synthesis claim, for immunization, is that using more information about community structure improves outbreak control: local strategies outperform global ones in networks with well-separated communities, while global strategies win when community structure is loose. If correct, the practical consequence is that intervention design should begin by measuring community strength rather than applying a default centrality measure.","feed_headline":"Immunization works better when it uses community structure","feed_subtitle":"Local strategies win in tight communities, global ones in loose networks; match the strategy to the modular strength.","key_machinery":"The organizing apparatus is a local-versus-global axis of node influence in modular networks, expressed through bridge nodes (nodes carrying inter-community links) and hub nodes (nodes carrying intra-community links). Its formal anchor is modularity $Q$, defined against the configuration model, together with the stochastic block model family; the equivalence between maximizing generalized modularity and maximum-likelihood inference of the degree-corrected planted partition model, with resolution parameter $\\gamma = (\\omega_1-\\omega_0)/(\\log\\omega_1-\\log\\omega_0)$, is what lets the paper treat detection quality and immunization choice as two sides of the same community-strength axis. In the immunization section, the load-bearing mechanism is that when few inter-community links exist, outbreaks remain local, so community hubs are the right targets; when many such links exist, bridges carry the outbreak globally.","core_discovery":"The paper's central claim, developed across its three sections, is that community structure is a quantitative, measurable property that should govern both how we detect groups and how we intervene in a network. On detection, it presents the stochastic block model as the principled generative foundation and reports that maximizing generalized modularity is equivalent to maximum-likelihood inference of the degree-corrected planted partition model, with a resolution parameter whose admissible range can be bounded. On dynamics, it argues that time-evolving communities can be recovered by snapshot matching, evolutionary algorithms, or incremental and online methods, each trading off accuracy, smoothness, and information. On immunization, the synthesis claim is that strategy quality rises with the amount of community-structure information used: local strategies outperform global strategies when communities are well separated, global strategies outperform local ones when communities are loose, combined strategies do best overall, and overlapping nodes act as epidemic carriers between modules.","pith_inferences":["A testable extension the authors do not develop: use community strength itself as a tunable parameter in a single adaptive strategy, so the same algorithm shifts weight from hubs to bridges as measured modularity decreases.","Because the surveyed rankings come from heterogeneous benchmarks, a fair comparison would require stratified benchmarks that vary only the ratio of inter- to intra-community links while holding degree distribution and size fixed; the paper's conditional claims predict that strategy rankings will invert across that axis.","The same local/global logic likely transfers to other diffusion processes on modular networks, such as misinformation or computer-virus spread, where intervention costs differ; the paper never makes this analogy explicit.","An adaptive stochastic strategy that estimates bridge-ness from short random walks could capture most of the benefit of deterministic strategies at a fraction of the information cost, which would make the semi-stochastic direction concrete."],"forward_implications":["If more community information improves immunization, deterministic strategies with full network knowledge should generally beat stochastic ones, and stochastic strategies should be redesigned to estimate community structure locally.","In networks with strong community structure, prioritize local hubs or community core nodes; in loose networks, prioritize bridge nodes; combination strategies that score both dimensions should be the safest default.","Overlapping nodes are high-value targets: membership-based and overlap-aware strategies can outperform degree, betweenness, and coreness in dense modular networks.","Modular centrality, or any centrality recast as local and global components, is a promising route because it is agnostic to the base centrality and leaves room for tuning the combination.","Future work should aim at semi-stochastic strategies that sit between fully local random-walk methods and fully global ranking methods."],"supporting_citations":[{"why":"Proves that modularity maximization is equivalent to maximum-likelihood inference for the degree-corrected planted partition model, grounding the detection half of the argument.","marker":"[5]"},{"why":"Supplies asymptotic upper and lower bounds on the generalized modularity resolution parameter and a progressive agglomerative algorithm, addressing the resolution limit.","marker":"[6]"},{"why":"Introduces the Community Bridge Finder strategy and provides the premise that community structure shapes epidemic dynamics.","marker":"[11]"},{"why":"Introduces the Bridge-Hub Detector strategy, a central stochastic method whose reported performance anchors the immunization comparisons.","marker":"[12]"},{"why":"Defines the degree-corrected stochastic block model, the core generative model used throughout the detection section.","marker":"[19]"},{"why":"Establishes the resolution limit of modularity-based community detection, motivating the search for bounded resolution parameters.","marker":"[25]"},{"why":"Introduces the random-walk overlap selection strategy, the main stochastic method for overlapping communities.","marker":"[90]"},{"why":"Proposes the Module-based global immunization strategy on the coarse-grained community network, a key global-strategy baseline.","marker":"[93]"},{"why":"Defines modular centrality as a two-dimensional local/global measure, formalizing the synthesis claim about combining both influence types.","marker":"[108]"}],"fun_headline_variants":["Community structure steers both detection and immunization","Best immunization strategy depends on community tightness","Stochastic block model sharpens community detection","Match immunization to community strength for better control","Communities shape how to detect and defend networks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The comparative conclusions about immunization assume that the SIR and SI simulations and synthetic benchmarks used across the many cited studies are consistent with each other and represent real-world contact networks, even though the review itself runs no unified benchmark.","fun_headline_variants_meta":{"raw":{"variants":["Community structure steers both detection and immunization","Best immunization strategy depends on community tightness","Stochastic block model sharpens community detection","Match immunization to community strength for better control","Communities shape how to detect and defend networks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000283,"raw_usage":{"total_tokens":1696,"prompt_tokens":995,"completion_tokens":701,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":611,"completion_tokens_details":{"reasoning_tokens":634}},"tokens_in":611,"tokens_out":701,"duration_ms":7587,"temperature":1.0,"reasoning_tokens":634,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:28:20.391912+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a single controlled benchmark on networks that differ only in community strength, holding degree sequence, size, and epidemic parameters fixed, and compare one local strategy with one global strategy. If the global strategy wins in strongly modular networks, or the local strategy wins in loosely modular networks, the paper's central conditional ranking is falsified; observing no monotone relationship between added community information and epidemic size would also undercut it.","supporting_citations":[{"cited_title":"Asymptotic resolution bounds of generalized modularity and multi-scale community detection","cited_arxiv_id":"1902.04243","evidence_quote":"Supplies asymptotic upper and lower bounds on the generalized modularity resolution parameter and a progressive agglomerative algorithm, addressing the resolution limit."},{"cited_title":"Taghavian, M","cited_arxiv_id":null,"evidence_quote":"Introduces the random-walk overlap selection strategy, the main stochastic method for overlapping communities."},{"cited_title":"Masuda, Immunization of networks with community structure, New Journal of Physics 11(12), 123018 (2009)","cited_arxiv_id":null,"evidence_quote":"Proposes the Module-based global immunization strategy on the coarse-grained community network, a key global-strategy baseline."},{"cited_title":"Ghalmane, M","cited_arxiv_id":null,"evidence_quote":"Defines modular centrality as a two-dimensional local/global measure, formalizing the synthesis claim about combining both influence types."}],"review_version":1}