{"id":"7ea371e1-5035-4555-bf5b-0a22b5a0b7dd","arxiv_id":"2510.24360","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Overlapping nodes that belong to multiple circles show persistently higher influence-spreading centrality than non-overlapping nodes across four real social networks under both simple and complex contagion models.","lead":"A study of social networks finds that nodes belonging to multiple circles, such as hobby and work groups, spread influence more effectively than nodes in a single circle. The finding could guide targeted marketing, misinformation containment, and immunization strategies by identifying high-impact individuals from group memberships alone.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim that overlap itself drives influence is not yet supported: OL and NOL groups differ systematically in degree, and all reported comparisons are unadjusted for degree.","rationale":"The reader's weakest assumption identifies exactly the concern I consider most load-bearing: the OL/NOL comparison is unadjusted for degree, so the observed influence advantage may be confounded. I read the paper in good faith: it is a clear empirical study, uses public data, includes useful sensitivity analyses for edge weights and circle sizes, and does not overstate its mechanistic claims in the Discussion. However, the abstract's 'consistently exhibit greater influence' and the strategic-implications framing require that overlap itself carries explanatory weight beyond connectivity. The paper does not provide that evidence. Because the missing analysis is straightforward and the comparative claim may survive degree matching, this is a CONDITIONAL situation rather than a rejection. I therefore keep the reader's verdict unchanged, with the concrete degree-matched check as the condition.","tokens_in":13461,"tokens_out":3165,"duration_ms":33134,"concrete_test":"Recompute the temporal relative differences and geometric-mean ratios with degree-matched samples. For each network, for each OL node, draw a NOL node with identical degree (or nearest degree within ±1, with replacement) from the same network; recompute Eqs. 3, 8, and 10 restricted to the matched OL and NOL sets, for both SC and CC at the same T values used in Figs. 4–8. If the matched curves and R-ratios collapse toward 1, the headline claim is an artifact of degree; if a substantial gap remains, overlap has an independent association.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim—that overlapping nodes 'consistently exhibit greater influence than non-overlapping ones'—rests entirely on comparisons of mean and geometric-mean centralities between OL and NOL groups (Eqs. 3, 8–10; Figs. 4–8) without controlling for node degree. This is load-bearing because circle membership is not degree-neutral: nodes in multiple circles typically have more social circles and thus more edges, and Out-centrality and Betweenness Centrality derived from the Influence Spreading Matrix are strongly driven by the number and length of paths emanating from a node. The reported 90% higher Out-centrality in LJ/ORK and the R-ratios exceeding one in every network may therefore reflect the fact that OL nodes are, on average, better connected, rather than an independent effect of overlap itself. The paper's own Appendix B is suggestive here: when edge weights are high, transmission probability approaches one and the OL advantage diminishes, which is what a connectivity-based mechanism would predict. Additionally, Figure 2 shows FB OL nodes shifted toward lower centrality in the top decile, so the 'consistently' claim is already qualified even before matching. Without a degree-matched or degree-stratified comparison, the data cannot distinguish 'overlap matters' from 'overlap is a proxy for degree.' This does not invalidate the descriptive result, but it does undermine the stronger topological/strategic interpretation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the role of nodes that belong to multiple overlapping circle structures in influence spreading on networks. Using a probabilistic Influence Spreading Model (ISM), the authors define three centrality metrics—In-centrality, Out-centrality, and Betweenness Centrality—and compare their values between overlapping (OL) and non-overlapping (NOL) nodes in four real-world network datasets (Facebook, LiveJournal, Orkut, and Wikipedia categories) under both simple and complex contagion. The main reported finding is that OL nodes exhibit consistently higher Out-centrality and Betweenness Centrality than NOL nodes, with relative differences up to 90% in LiveJournal and Orkut, and geometric-mean ratios exceeding one in all networks. The paper also analyzes how the minimum circle size affects the OL advantage, concluding that influential nodes reside predominantly in larger circles.","tokens_in":13741,"tokens_out":3549,"duration_ms":34465,"significance":"If the central claim were established, the paper would contribute to understanding the role of overlapping substructures in spreading dynamics and would have practical implications for targeted intervention or influence maximization. The study has notable strengths: it uses externally defined ground-truth circles rather than algorithmically detected communities, compares multiple networks with diverse structural properties, tests both simple and complex contagion, and provides bootstrap confidence intervals and a robustness check over edge weights. However, the current analysis does not rule out the simpler explanation that the observed OL advantage is a proxy for higher node degree, and the reliance on the ISM model—referenced but not described in sufficient detail—limits the reader's ability to assess the generality of the findings. The paper's headline claim is also stronger than the presented evidence: the FB dataset and the In-centrality results show only weak or no consistent advantage.","major_comments":[{"comment":"The comparison between OL and NOL nodes is not adjusted for node degree. Since nodes in multiple circles are likely to have more social connections, and since Out-centrality and Betweenness Centrality derived from the ISM are strongly influenced by the number and length of paths emanating from a node, the reported 90% higher Out-centrality (LJ, ORK) and the geometric-mean ratios R>1 (Figs. 7–8) may reflect the fact that OL nodes are better connected rather than an effect of overlap itself. Appendix B reinforces this concern: when edge weights approach 1, the OL advantage diminishes, consistent with a connectivity-driven mechanism. The authors should provide a degree-matched or degree-stratified comparison (e.g., matching each OL node to a NOL node of the same degree, or including degree as a covariate) to support the claim that overlap has an independent effect on influence.","section":"Results, Eqs. (3), (8), (10), Figs. 4–8"},{"comment":"The Influence Spreading Model is only referenced (ref. 30) rather than described; the reader is told that the model outputs a matrix C of pairwise influence probabilities, but not the update equations, the handling of simple versus complex contagion, or the role of parameters such as the uniform edge weight 0.05 and maximum path length 100. Because all centrality metrics and the betweenness measure are defined on this ISM, the reported results are entirely internal to this model. The authors should either summarize the model's equations in the manuscript or provide a more detailed description in an appendix, and they should discuss how the model's parameters are chosen and whether the qualitative findings are robust to reasonable variations. Ideally, the model should be validated against an external spreading dataset or at least compared with a standard SIR-like simulation to ensure the centralities are not an artifact of the ISM's specific formulation.","section":"Methods, 'Methods' subsection"},{"comment":"The abstract states that 'at each stage of the spreading process the overlapping nodes consistently exhibit greater influence than the non-overlapping ones,' but this is not fully supported by the authors' own results. In the FB dataset, Figure 2 shows OL nodes shifted to lower Betweenness Centrality in the top decile, and the text reports only a 'smaller difference' for FB. Moreover, the In-centrality relative difference decreases smoothly over time (Fig. 4b) and the geometric-mean ratio for In-centrality concentrates around unity (Fig. 7c). The strong claim of consistency should be qualified to specify the metrics (Out-centrality and Betweenness Centrality) and datasets for which the effect is robust, or the analysis should be extended to establish a consistent effect across all three metrics.","section":"Abstract; Results, Fig. 2, Fig. 7c"}],"minor_comments":[{"comment":"There is a typo: 'importanc' should be 'importance'.","section":"Abstract"},{"comment":"The parameter choice of edge weight 0.05 is defended in Appendix B, but the discussion of weights appears after the results; consider moving a brief justification of the edge weight and maximum path length to the Methods section for readers who do not read the appendix.","section":"Methods, 'Choosing the Edge Weights' (Appendix B)"},{"comment":"The caption says 'Cumulative density' but the figure plots cumulative distribution functions; the wording should be corrected to 'cumulative distribution function'.","section":"Results, Fig. 2 caption"},{"comment":"The sentence 'Subsequently, only few very peripherial and isolated nodes would be classified as NOL' contains a typo: 'peripherial' should be 'peripheral'.","section":"Discussion, 'The Choice of Circles'"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a descriptive empirical study that is largely consistent with prior work by the same group (ref. 32) and with existing literature on overlapping communities. The main concern is not novelty but whether the central claim survives a degree-controlled analysis. If the authors can provide a degree-matched comparison and a more self-contained description of the ISM, the paper could become acceptable. The fit with a general computational social science journal is reasonable, though the contribution is somewhat incremental."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short take: this is a solid empirical paper with a clear protocol and a genuinely good negative control, but the load-bearing claim is undercut by the lack of any degree-matching. The descriptive patterns are probably real; the interpretation that overlap is doing the causal work is not yet supported.\n\nWhat's new: the time-resolved In/Out/Betweenness gaps, the circle-size dependence in LJ, and the Pokec synthetic-circles null result. The null result was a good idea—it shows that arbitrary attribute-based circles don't reproduce the effect, which strengthens the case that real circle structure matters. The bootstrap confidence intervals on the geometric mean ratios are a nice touch.\n\nWhere it wobbles: every core comparison in Figures 4–8 compares centralities of OL vs NOL groups without controlling for degree. Circle membership is not degree-neutral: nodes in many circles tend to have more edges, and path-based metrics like Out-centrality and the ISM-based betweenness will naturally favor higher-degree nodes. Appendix B even shows the OL advantage shrinks as edge weights approach 1, which is exactly what a connectivity-driven mechanism would predict. The FB top-decile reversal in Figure 2 also qualifies the 'consistently' language in the abstract. The authors are aware of some of these issues in the Discussion, but the abstract still overstates. Separately, the ISM is from the authors' own prior work, referenced rather than described, and not validated against any external spreading data. That makes the whole result internal to one modeling framework—not a fatal flaw, but it limits how much I trust the quantitative claims.\n\nSo the central argument as stated doesn't fully hold up. The distributional differences are descriptive and worth reporting, but the strategic interpretation requires showing that overlap adds predictive power beyond degree. That's fixable: a degree-stratified or matched comparison, or a model that includes degree as a covariate, would answer the main question directly.\n\nWho this is for: people working on influence maximization and targeted immunization who want a careful look at whether overlap can serve as a cheap signal. The paper deserves a serious referee, but it needs a major revision before I'd trust the conclusions. I'd send it forward, with the request that the authors run a degree-controlled analysis and dial the abstract back to match what the data actually show.","headline":"Useful descriptive study of overlap and spreading centrality, but the missing degree-matched baseline means the paper cannot yet support the claim that overlap itself drives influence.","tokens_in":14229,"tokens_out":2433,"would_cite":false,"duration_ms":23369,"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":"Nodes belonging to multiple overlapping circles consistently show higher influence in spreading processes than nodes in a single circle, under both simple and complex contagion, across four real-world networks.","keywords":["overlapping nodes","influence spreading","complex contagion","simple contagion","centrality","influence spreading matrix","circles","social networks"],"falsifier":"Repeat the analysis after matching each overlapping node to a non-overlapping node with the same degree, or re-running the simulations on degree-preserving randomizations of circle memberships; if the Out-Centrality advantage and the geometric-mean ratios above one vanish, then overlap per se does not drive the effect.","tokens_in":13286,"feed_emoji":"🕸️","tokens_out":7378,"duration_ms":64478,"temperature":0.7,"pith_summary":"This paper asks whether nodes that sit in several overlapping network circles matter more for spreading influence than nodes that belong to only one. Using a probabilistic influence spreading model on real ego-networks from Facebook, LiveJournal, Orkut, and Wikipedia, the authors compare overlapping and non-overlapping nodes with three centrality measures: In-Centrality (susceptibility), Out-Centrality (spreading power), and Betweenness Centrality (mediating role). They find that overlapping nodes are consistently more influential at every stage of the process, with Out-Centrality differences reaching about 90% in the saturated phase for LiveJournal and Orkut, and geometric-mean ratios exceeding one for Out- and Betweenness Centrality in every network under both contagion mechanisms. The result matters because it identifies a small, recognizable class of nodes that could act as levers for accelerating or containing spread, and it also clarifies the distinction between local, attribute-driven circles and global community structures.","feed_headline":"Nodes in multiple circles out-spread single-circle nodes","feed_subtitle":"Network model finds overlap boosts influence in both simple and complex contagion.","key_machinery":"The central object is the Influence Spreading Matrix (ISM), a matrix $C$ whose entry $C_{ij}$ is the probability that influence originating at node $i$ reaches node $j$ under the probabilistic Influence Spreading Model. From this matrix the paper derives three node-level metrics: In-Centrality (column sums, susceptibility), Out-Centrality (row sums, spreading power), and an ISM-based Betweenness Centrality defined as the relative decrease in total network cohesion when the node is removed. Because the ISM accounts for all propagation paths rather than only shortest paths, these metrics capture probabilistic and temporal features of spreading that conventional centrality measures miss, and the same framework accommodates both simple contagion (single-pass self-avoiding paths) and complex contagion (recurrent interactions and feedback). The key comparison is the relative difference between average metric values for overlapping and non-overlapping nodes, backed by bootstrap geometric-mean ratios.","core_discovery":"The paper's central claim is that overlapping nodes are not peripheral participants but consistently the strongest drivers of influence spreading. This is demonstrated by computing, from the Influence Spreading Matrix, the average In-, Out-, and Betweenness centrality of the two node classes and comparing them over time. At the start of spreading, overlapping nodes show markedly higher In-Centrality, indicating greater exposure; as the process saturates, the In-Centrality gap narrows while the Out-Centrality gap persists or grows, so overlapping nodes keep spreading power long after the initial wave. The betweenness analysis shows that overlapping nodes retain their mediatory role over a longer period, and bootstrap ratios of geometric means confirm the effect in every dataset and under both contagion models, with the exception of a less decisive In-Centrality ratio hovering near unity. The authors further report that the circle definition matters: when only the largest circles are kept, the overlap advantage shrinks only gradually, implying that the most influential overlapping nodes live in large circles rather than small triads.","pith_inferences":["A degree-matched replication would show whether the overlap advantage is independent of connectivity; the paper does not perform this matching.","The same ISM-based machinery could be applied to directed and temporal networks, where path reversal symmetry breaks and In- versus Out-Centrality differences become more informative.","The concentration of super-influencers in large circles suggests a practical two-step targeting heuristic: find large circles, then pick nodes that belong to several, which could be tested against standard influence-maximization algorithms.","Because the bridging mechanism is largely topological, overlapping communities rather than circles should show an even stronger effect, a prediction the authors leave open."],"forward_implications":["Immunization and targeted influence campaigns can concentrate on overlapping nodes, since these nodes keep a disproportionate spreading role even after saturation.","Out-Centrality computed from the ISM offers a computationally cheaper proxy for Betweenness Centrality when only relative differences between node classes are needed.","Circle-definition choices change the measured overlap effect, so any comparison across datasets must state the minimum circle size; restricting to large circles preserves the influence advantage and locates key influencers in the largest circles.","The ordering of overlapping over non-overlapping nodes holds under both simple and complex contagion in all four networks, so the effect appears robust to the contagion mechanism.","Overlap alone is not enough to identify true influencers; the most influential nodes likely sit at the intersection of circles and community structures."],"supporting_citations":[{"why":"Introduces the probabilistic Influence Spreading Model and the Influence Spreading Matrix that underlies all three centrality measures.","marker":"30"},{"why":"Defines the ISM-based Betweenness Centrality and reports the similarity between Out-Centrality and betweenness trends.","marker":"32"},{"why":"Provides the ego-Facebook ground-truth circles used as one of the four datasets and motivates the circle-based overlap definition.","marker":"5"},{"why":"Presents prior evidence that overlapping nodes act as bridges and principal drivers of contagion, providing the comparison baseline the paper extends.","marker":"12"},{"why":"Shows that overlapping-node immunization reduces epidemic prevalence, motivating the strategic relevance of the paper's results.","marker":"17"},{"why":"Reports that the top centrality quartile contains more overlapping nodes, the distribution-level result the paper refines with percentile analysis.","marker":"25"},{"why":"Establishes overlapping community structures in complex networks and underpins the overlap concept used throughout.","marker":"10"}],"fun_headline_variants":["Overlapping nodes consistently show greater spreading power","Influence spreading favors nodes in multiple circles","Circle overlap enhances node influence in networks","Overlapping nodes excel at simple and complex contagion","Large circles boost the effect of overlapping nodes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The comparison does not control for node degree or other attributes, so the entire case rests on the assumption that the overlap itself, rather than the higher connectivity that often comes with it, is what produces the measured influence advantage.","fun_headline_variants_meta":{"raw":{"variants":["Overlapping nodes consistently show greater spreading power","Influence spreading favors nodes in multiple circles","Circle overlap enhances node influence in networks","Overlapping nodes excel at simple and complex contagion","Large circles boost the effect of overlapping nodes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000305,"raw_usage":{"total_tokens":1743,"prompt_tokens":929,"completion_tokens":814,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":545,"completion_tokens_details":{"reasoning_tokens":747}},"tokens_in":545,"tokens_out":814,"duration_ms":7222,"temperature":1.0,"reasoning_tokens":747,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:41:19.896759+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the analysis after matching each overlapping node to a non-overlapping node with the same degree, or re-running the simulations on degree-preserving randomizations of circle memberships; if the Out-Centrality advantage and the geometric-mean ratios above one vanish, then overlap per se does not drive the effect.","supporting_citations":[{"cited_title":"Influence spreading model used to analyse social networks and detect sub-communities.Comput","cited_arxiv_id":null,"evidence_quote":"Introduces the probabilistic Influence Spreading Model and the Influence Spreading Matrix that underlies all three centrality measures."},{"cited_title":"& Kaski, K","cited_arxiv_id":null,"evidence_quote":"Defines the ISM-based Betweenness Centrality and reports the similarity between Out-Centrality and betweenness trends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Presents prior evidence that overlapping nodes act as bridges and principal drivers of contagion, providing the comparison baseline the paper extends."},{"cited_title":"& Cherifi, H","cited_arxiv_id":null,"evidence_quote":"Shows that overlapping-node immunization reduces epidemic prevalence, motivating the strategic relevance of the paper's results."},{"cited_title":"K., Ali, W","cited_arxiv_id":null,"evidence_quote":"Reports that the top centrality quartile contains more overlapping nodes, the distribution-level result the paper refines with percentile analysis."}],"review_version":2}