{"id":"aa6f2ded-5bf1-41da-aa6a-c5df0bdf0ff2","arxiv_id":"2506.02669","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"On growing scale-free networks with homophily, bounded-confidence opinion dynamics produce sudden merges of minor opinion clusters into major ones, triggered by arriving 'bridge' agents who connect both clusters.","lead":"This paper simulates how opinions spread on social networks that grow as new people join, and finds that small opinion groups can suddenly collapse into a big group, which the authors call an 'opinion cascade'. The result matters because it offers a mechanism for how minority views get absorbed quickly online, which could be relevant for understanding misinformation and polarization.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theoretical explanation of cascades hinges on Eq. (3)-(4), which assume each initialized cluster has total degree 2m n_i; if cluster construction is not specified, the claimed bridge-probability fit at t=0 in Fig. 3 may be an artifact of an arbitrary internal topology.","rationale":"The paper's central new observation is the occurrence of opinion cascades in a growing, homophilous scale-free network, with bridge agents as the proposed trigger. The simulation evidence in Fig. 1 is suggestive, and the controlled 10,000-replica experiments in Section 4.2 are a serious attempt to test the bridge mechanism. However, the quantitative support for that mechanism is the theoretical bridge probability derived in Section 4.1, and that derivation rests on Eqs. (3)-(4) equating the total degree in each initial cluster to 2m n_i. The paper never describes how the clusters in Section 4.2 are constructed. If they were generated by the model's own arrival process, Eqs. (3)-(4) are a reasonable approximation; if they were initialized as complete graphs or some other topology, the derivation is off by a large factor. The reader's phrasing that each agent has degree m on average slightly misstates the implication of Eq. (3) (average degree is 2m), but the underlying worry about unspecified internal topology is correct. This is not a cosmetic issue: the bridge probability enters the explanation of why beta approximately 15 cascades are fast while beta approximately 30 cascades are slow (Figs. 3-5), so an unjustified degree assumption makes the theoretical validation construction-dependent. My proposed test isolates this by varying only internal topology at t=0. If the concern lands, the paper's explanation is still plausible but requires a stated cluster-generation rule or a revised mean-field approximation. The reader's conditional verdict is therefore the right one, and my read does not change it.","tokens_in":7801,"tokens_out":19219,"duration_ms":182620,"concrete_test":"Run the controlled experiment of Section 4.2 twice, changing only the internal topology of the two initial clusters: (A) grow each cluster by sequential arrivals with m=3 internal links so that Eqs. (3)-(4) hold approximately; (B) initialize each cluster as a complete graph. Keep n1=225, n2=25, theta1=0.5, theta2=0.83, epsilon=0.3, delta=1, and beta=15. At t=0, measure the empirical probability that the first arriving agent connects to both clusters, averaged over 10,000 replicas, and compare with Eq. (7). If case B deviates from Eq. (7) substantially (e.g., >20% relative error) while case A matches, the manuscript must specify and justify the cluster construction before the theoretical explanation can be accepted; if both match, the degree assumption is robust to topology and the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4.1 derives the probability that an arriving agent bridges the major cluster C1 and minor cluster C2. The key step is Eqs. (3)-(4): sum_{i in C1} k_i = 2m n1 and sum_{i in C2} k_i = 2m n2, justified by the arrival of an agent connected to agents of C1 only creating 2m oriented connections. This is only true if each cluster was itself assembled by the same growth process, adding agents one at a time with m internal links per arrival (up to a small seed correction). It is not true for an arbitrary connected-only-to-this-cluster initialization. Section 4.2 does not specify the internal topology of the n1=225 and n2=25 clusters. If they are complete graphs, the degree sums are n1(n1-1)=50,400 and n2(n2-1)=600 instead of 1,350 and 150 for m=3, changing Eq. (7) by a large factor. The paper reports that the theoretical bridge probability fits experimental frequencies quite very accurately at t=0; without the cluster-construction details, that fit cannot be distinguished from a match to the simulator's hidden construction. Because this bridge probability is the quantitative basis for the paper's causal claim that cascades are triggered by bridge agents, the theoretical validation is conditional on an unstated and potentially wrong degree assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies a growing scale-free network model in which new agents attach with probability proportional to degree times an exponential homophily term in opinion distance, while connected agents interact via the bounded-confidence rule. It reports that for intermediate values of the homophily parameter beta, minority opinion clusters form at the periphery of opinion space, persist for a time, and then abruptly merge with the majority cluster, a process the authors call 'opinion cascades.' To explain this, the authors derive an approximate probability that a newly arriving agent connects to both the major and minor clusters (a 'bridge' agent) and validate this formula against simulations initialized with one large and one small cluster. They argue that the faster growth of bridge agents for lower beta triggers the cascade, while for higher beta the bridge population grows too slowly and the merging is smoother.","tokens_in":8096,"tokens_out":5817,"duration_ms":60692,"significance":"If the reported phenomenon is robust, the paper is a useful contribution: it identifies a plausible mechanism (bridge agents) linking network growth, homophily, and bounded-confidence dynamics, with a qualitative prediction about merging speed as a function of beta. The manuscript has clear strengths: a precise model specification with pseudo-code, a large number of replicas (10,000) in the controlled experiments, and an explicit attempt to go beyond purely qualitative simulation results. However, the key quantitative assumption behind the theoretical derivation is currently unverified, and the validation in Figure 3 is partly a consistency check rather than an independent prediction. The central causal claim is therefore not yet established at the level the paper claims, though it is likely fixable with additional specification and testing.","major_comments":[{"comment":"The identities sum_{i in C1} k_i = 2m n1 and sum_{i in C2} k_i = 2m n2 are asserted after noting that an arriving agent connected only to one cluster creates 2m oriented connections. This is only correct if each cluster was assembled by the same growth rule, i.e., each added node links to m existing nodes. Section 4.2 does not specify how the initial clusters of 225 and 25 agents are wired internally. If they are fully connected, as the initial network of N0 agents is described in Section 2, the degree sums are n1(n1-1) = 50,400 and n2(n2-1) = 600 rather than 1,350 and 150 for m=3, changing Eq. (7) by a large factor. Because this assumption enters the theoretical curves in Figure 3, the reported fit at t=0 cannot be evaluated without knowing the cluster construction.","section":"§4.1, Eqs. (3)-(4)"},{"comment":"The theoretical probabilities are computed using the average cluster positions and sizes from the same 10,000 replicas that produce the empirical frequencies. This makes Figure 3 a consistency check rather than an independent test of the bridge-agent mechanism. To support the claim that bridge agents are the cause of cascades, the authors should either measure the degree sums directly in the simulations and insert them into Eq. (2), or derive the degree sums from a fully specified construction and then compare the resulting formula with data not used in its calibration.","section":"§4.2, Fig. 3"},{"comment":"The central empirical distinction between sudden cascades for beta in [15,20] and smoother merging for beta > 20 is illustrated by two representative runs (beta=15 and beta=30) without error bars or a systematic parameter sweep. Since the phenomenon is stochastic and the paper emphasizes the shape of the merging process, the reader needs either ensemble-averaged trajectories with confidence bands or a quantitative measure of cascade suddenness across a range of beta values.","section":"§3.1, Fig. 1"},{"comment":"The product formulas for the probability of connecting only to C1 or only to C2 account for the decreasing number of available agents in the chosen cluster but ignore that each accepted link increases the degree, and hence the future attachment weight, of the selected agent. For m=3 this may be a minor correction, but the derivation as written is not the exact probability under the model's sequential attachment rule; the approximation should be stated explicitly.","section":"§4.1, Eqs. (9)-(10)"}],"minor_comments":[{"comment":"The condition 'if random(0, 1) < 1/N' appears to omit the parameter delta; the model description and the input list give the probability of adding a new agent as delta/N.","section":"Algorithm 1, line 8"},{"comment":"There are several typos and awkward phrasings, including 'quite very accurately,' 'frequence' in §3.1, and 'probability to of arriving bridge agent' in §4.2.","section":"§4.2"},{"comment":"The panels would benefit from error bars or shaded confidence intervals, especially because the text claims that the fit is 'quite well' at t=50 and that accuracy 'decreases significantly' at t=100.","section":"Fig. 3"},{"comment":"Inserting average positions and sizes into a nonlinear formula can bias the theoretical curves; a more direct comparison would compute P_b for each replica and then average the resulting probabilities.","section":"§4.2"},{"comment":"The effect of the parameters delta and m on the presence and sharpness of cascades is not systematically explored, although Fig. 2 uses delta=6 while Fig. 1 uses delta=3; a sentence on parameter sensitivity would help.","section":"§2"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the journal's scope and the observed phenomenon is potentially interesting. My main reservation is that the theoretical validation is currently closer to a consistency check than an independent confirmation, and the unverified degree-sum assumption in Eqs. (3)-(4) could invalidate the quantitative fit in Fig. 3. I would be satisfied by a revision that specifies the cluster construction, measures or derives the degree sums, and adds an out-of-sample or simulation-free test of the bridge-probability formula."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing here is the observation that on growing scale-free networks with bounded-confidence interactions and homophily-based link formation, minor opinion clusters can sit on the periphery for a while and then suddenly merge with a major cluster. That is a clean, reproducible-in-principle simulation result, and it is distinct from the authors' earlier growing-fully-connected work. The paper is well written and the model is described clearly enough to re-implement.\n\nWhat it does well: the bridge-agent mechanism is plausible, and the comparison between β=15 and β=30 in Figs. 3–5 gives a coherent qualitative explanation for why merging is fast in one case and slow in the other. The paper is honest about the approximations in the theory.\n\nSoft spots: the main one is Eqs. (3)–(4). The paper claims that for a cluster assembled by the same growth process, total oriented degree is 2m n_i. But the initialized clusters in §4.2 are not described as BA-grown; they are just 'connected only to agents of this cluster.' If they are complete graphs, the degree sum is n_i(n_i−1), which changes the bridge probability by a large factor. The fact that the theoretical curves fit 'quite very accurately' at t=0 then becomes less impressive: it could be matching the simulator's hidden construction rather than validating a derivation. Second, the validation is a consistency check: theoretical curves are computed using simulated cluster positions and sizes, then compared to simulated frequencies. That is not an independent prediction. Third, no error bars on the main simulation figures and no systematic parameter sweep; we don't know how robust the cascade window (β ~ 15–20) is. Minor: no code or data released.\n\nProportion: the central observation — cascades happen — does not depend on the theory. I believe it. The theory is an attempt to explain the speed difference, and the qualitative story (bridges become more probable as clusters approach) is sound even if the quantitative fit is overclaimed. The degree-sum issue is a real gap in the derivation as written, but it is fixable by specifying cluster initialization or deriving the degree sum for whatever initialization is used.\n\nWho this is for: people working on opinion dynamics, network growth, and minority influence. It is a within-subfield contribution, not a paradigm shift.\n\nRecommendation: I would send it to peer review, with the expectation that the authors must clarify the cluster construction, show error bars, and either provide an independent prediction or soften the claim that the theory is validated. The cascade phenomenon itself deserves to be in the literature.","headline":"New simulation result — cascades on growing scale-free networks with homophily — looks real, but the theoretical fit in §4 rests on an unstated cluster-degree assumption and a same-data consistency check.","tokens_in":8591,"tokens_out":1952,"would_cite":true,"duration_ms":18599,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"In growing social networks, bridge agents are what make minority opinions suddenly merge with the majority.","keywords":["bounded confidence","opinion dynamics","scale-free networks","opinion cascades","homophily","bridge agents","growing networks"],"falsifier":"Run the two-cluster experiment with a specified internal topology (e.g., fully connected clusters or Erdős–Rényi with average degree m) and measure the time until the minor cluster's average opinion crosses halfway to the major cluster; if this merging time differs substantially from the one predicted by the 2 m n_i degree assumption, the bridge-probability explanation is falsified. A simpler check is to compute the empirical average degree inside each cluster just before cascade events and see whether it is close to m.","tokens_in":7586,"feed_emoji":"🌊","tokens_out":6342,"duration_ms":58110,"temperature":0.7,"pith_summary":"Bounded-confidence opinion dynamics on a growing, scale-free network with homophily can produce 'opinion cascades': small opinion clusters that form on the periphery of opinion space, stay stable for a time, and then abruptly merge with the dominant cluster. The paper establishes that this merging is triggered by newly arriving agents that, by chance, link to both clusters at once; these bridge agents then pull the clusters' opinions together through repeated interactions. The authors derive an approximate formula for the probability that an arriving agent is a bridge, and show that its time evolution explains why merging is fast at moderate homophily and slow at strong homophily. The result matters because it gives a mechanistic explanation for sudden opinion shifts and for the absorption of minority groups into majorities, without relying on contagion or epidemic spreading.","feed_headline":"Bridge agents trigger sudden opinion cascades in growing networks","feed_subtitle":"Simulations show minority opinion clusters abruptly merge with the majority when a new agent links both.","key_machinery":"The central object is the link-choice probability φ_N(i) = k_i exp(−β|θ_N − θ_i|) / Σ_j k_j exp(−β|θ_N − θ_j|), which sets the chance that an arriving agent connects to each pre-existing node as a trade-off between degree k_i and opinion proximity. Using the assumption that each cluster's total degree is 2 m n_i, the paper reduces this to a two-cluster bridge probability P_2(n_1,n_2) = 1/(1 + (n_1/n_2) exp(β(θ_1+θ_2−2x))), then to P_b = 1 − P_{1,m} − P_{2,m}, the probability that a new agent of opinion x links to both the major and the minor cluster. This bridge probability is what explains the cascade: its growth in time, combined with the small number of links m, determines how fast the minor cluster's average opinion is pulled toward the major cluster.","core_discovery":"The paper claims that in the parameter range where both node degree and opinion proximity influence link formation (intermediate values of the homophily parameter β), the pairwise bounded confidence model on a growing Barabási–Albert-like network generates opinion cascades. Minor opinion clusters continually appear at distances greater than the confidence bound ε from a major cluster, remain stable for some time, and then suddenly merge with the major cluster when an arriving agent links to both clusters. Using a simplified two-cluster setup, the authors derive the probability P_b that a new agent becomes such a bridge, and they validate the qualitative form of this probability in simulations with 10,000 replicas. The comparison shows that for β=15 the bridge probability grows rapidly as clusters approach, making the number of bridges soon exceed the number of agents attached only to the minor cluster, which drives the fast merging observed at β=15; for β=30 the bridge probability stays low, so the minor cluster is slowly absorbed instead. The paper also notes that the cascade mechanism requires the number of new links per agent m to be small, since a large m lets bridge agents interact almost exclusively with the major cluster.","pith_inferences":["We extend the paper's bridge mechanism to conjecture that cascade speed is controlled by the ratio n_1/n_2 as well as β, and that the critical bridge probability at which runaway merging begins could be estimated analytically; this is not tested in the paper.","A testable extension: in a version of the model where arriving agents are drawn from a non-uniform opinion distribution (e.g., heavy-tailed), the cascade frequency should shift toward the peak of the arrival distribution, since bridge agents are more likely where arrivals concentrate.","The persistence of Louvain communities in the merged cluster suggests that even after the opinions have converged, the network's latent community structure may preserve a 'memory' of the pre-cascade division; this is our reading of the simulation result, not a claim the paper tests.","The mechanism may generalize to other network-growing rules: any link-formation process that mixes popularity with similarity could produce cascades analogous to these, since the only ingredient needed is the existence of agents that connect two otherwise isolated opinion basins."],"forward_implications":["For intermediate homophily strengths (β roughly 15–20) the model predicts that minor opinion groups with a confidence bound equal to that of the majority will be absorbed suddenly, whereas for stronger homophily (β > 30) the absorption is slow and gradual.","The mechanism requires a small number of links per newcomer m; when m is large, bridge agents interact mostly with the large cluster and minor clusters stay stable, matching the earlier fully-connected population result.","The theoretical bridge probability, although derived under strong simplifications, correctly reproduces the qualitative difference in merging speed between β=15 and β=30, giving a testable handle on when opinion cascades should occur.","If these cascades are real features of opinion dynamics, then polls or social-media signals should show stable peripheral opinions followed by sudden jumps toward the mainstream opinion, rather than smooth convergence.","The paper's conclusion suggests that minority groups can survive only by lowering their confidence bound, which the authors interpret as radicalisation."],"supporting_citations":[{"why":"Supplies the pairwise bounded-confidence interaction rule (Deffuant et al.) that the model implements.","marker":"[12]"},{"why":"Defines the bounded-confidence framework with a common confidence bound, used for the interaction and cluster structure.","marker":"[13]"},{"why":"Prior result on growing fully connected populations where minor clusters remain stable; the contrast motivates why the scale-free topology with bridges creates cascades.","marker":"[17]"},{"why":"Provides the homophily-dependent link probability φ_N that the model adopts.","marker":"[18]"},{"why":"Defines information cascades as massive sudden adoption, the phenomenon the paper argues it has found.","marker":"[1]"},{"why":"Establishes the baseline behavior of Deffuant bounded-confidence dynamics on Barabási–Albert networks, which the present growing-network setting extends.","marker":"[16]"}],"fun_headline_variants":["How bridge agents cause opinion cascades in growing scale-free networks","New links to both clusters spark sudden opinion merging","Opinion cascades arise when a single agent bridges minority and majority","Bounded confidence plus network growth yields abrupt opinion cascades","How one newcomer merges minority opinion into majority"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The predicted cascade speed rests on assuming each cluster has total degree 2 m n_i (every agent holding m connections on average) and that all agents in a cluster share one opinion, even though the simulations let clusters develop internal opinion spread and their internal connection structure is never specified.","fun_headline_variants_meta":{"raw":{"variants":["How bridge agents cause opinion cascades in growing scale-free networks","New links to both clusters spark sudden opinion merging","Opinion cascades arise when a single agent bridges minority and majority","Bounded confidence plus network growth yields abrupt opinion cascades","How one newcomer merges minority opinion into majority"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000627,"raw_usage":{"total_tokens":2873,"prompt_tokens":889,"completion_tokens":1984,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":505,"completion_tokens_details":{"reasoning_tokens":1904}},"tokens_in":505,"tokens_out":1984,"duration_ms":13352,"temperature":1.0,"reasoning_tokens":1904,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:18:10.365547+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the two-cluster experiment with a specified internal topology (e.g., fully connected clusters or Erdős–Rényi with average degree m) and measure the time until the minor cluster's average opinion crosses halfway to the major cluster; if this merging time differs substantially from the one predicted by the 2 m n_i degree assumption, the bridge-probability explanation is falsified. A simpler check is to compute the empirical average degree inside each cluster just before cascade events and see whether it is close to m.","supporting_citations":[{"cited_title":"Mixing beliefs among interacting agents,","cited_arxiv_id":null,"evidence_quote":"Supplies the pairwise bounded-confidence interaction rule (Deffuant et al.) that the model implements."},{"cited_title":"Opiniondynamicsandboundedconfidence:Models,analysis and simulation","cited_arxiv_id":null,"evidence_quote":"Defines the bounded-confidence framework with a common confidence bound, used for the interaction and cluster structure."},{"cited_title":"Bounded confidence model on growing populations,","cited_arxiv_id":null,"evidence_quote":"Prior result on growing fully connected populations where minor clusters remain stable; the contrast motivates why the scale-free topology with bridges creates cascades."},{"cited_title":"The role of homophily in the emergence of opinion controver- sies,","cited_arxiv_id":null,"evidence_quote":"Provides the homophily-dependent link probability φ_N that the model adopts."},{"cited_title":"Atheoryoffads,fashion,custom,andcultural change as informational cascades,","cited_arxiv_id":null,"evidence_quote":"Defines information cascades as massive sudden adoption, the phenomenon the paper argues it has found."},{"cited_title":"Simulation of consensus model of deffuant et al on a barabasi-albert network,","cited_arxiv_id":null,"evidence_quote":"Establishes the baseline behavior of Deffuant bounded-confidence dynamics on Barabási–Albert networks, which the present growing-network setting extends."}],"review_version":1}