{"id":"cef5cbb2-fa5d-4330-93d4-ac3b5cc519f3","arxiv_id":"2507.16046","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Using 29 million tweets, the study concludes that K-pop fans joined BLM activism mainly because they already shared the movement's beliefs, not because BTS led them, though the evidence is only moderate.","lead":"Why did K-pop fans join Black Lives Matter in 2020: shared beliefs or idol influence? Using tweets from about 2,900 users, this study concludes that belief alignment was the main driver, with BTS's statement amplifying rather than starting the activism.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The H1 evidence is post hoc: attractors 2 and 3 were selected after their spikes, and the two most mixed attractors (6 and 7) did not spike, so no statistical test distinguishes the belief-alignment claim from chance.","rationale":"The reader's conditional verdict correctly identifies the post hoc selection of attractors as a core weakness. My stress-test sharpens that concern: even if the Belief Landscape Framework is accepted on its own terms, the H1 inference is not statistically grounded, because the spiking attractors were chosen after the fact and no null model is offered. The paper's own limitation statement—'although coordinated spikes do not appear in the most heterogeneous attractors'—is an internal admission that the expected pattern was not cleanly observed. The H2 null result and RQ2 convergence finding are secondary; they do not rescue the central claim if H1 is unsupported. I do not see the BLF representational validity as independently load-bearing here, because the authors provide a coherence validation and a sensitivity analysis; the selection problem would remain even with a perfect belief representation. A permutation test of the spike-homogeneity association would settle whether the observed pattern is real. This reinforces rather than changes the reader's CONDITIONAL verdict, so I leave the verdict unchanged.","tokens_in":19736,"tokens_out":4519,"duration_ms":56828,"concrete_test":"Run a permutation test that preserves the pre-event attractor structure: for each of the 21 attractors, fit a null model of weekly K-pop and BLM activity from weeks before week 20 (e.g., negative binomial calibrated to pre-event counts). Simulate 10,000 null datasets with no Floyd event, apply the same z>2 event detector to weeks 20 and 21, and record the mean pre-event homogeneity rank of attractors showing coordinated spikes in both communities. Compare the observed mean rank (attractors 2 and 3: ranks 3 and 5) to this null distribution. If the observed rank is not below the 5th percentile, H1 is statistically indistinguishable from chance, and the abstract should be softened accordingly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that belief alignment, not opinion leadership, drove cross-cultural activism—rests on H1: coordinated spikes in mixed attractors before BTS's tweet. The paper identifies attractors 2 and 3 only after observing their spikes, then shows they rank 3rd and 5th in pre-event heterogeneity among 21 attractors (Table 1). This is a post hoc rank-based argument, not a test. The two most heterogeneous attractors, 6 and 7, did not spike, which the paper concedes. With 21 attractors, weekly z>2 event detection, and no multiple-testing correction, some coordinated spike pattern is expected by chance; a spike will always land in an attractor with some homogeneity rank. The analysis provides no permutation test, no comparison of spiking vs. non-spiking attractors, and no outcome-independent prediction. Even granting that the Belief Landscape Framework faithfully represents beliefs, the selection procedure prevents H1 from being evaluated. The paper itself says 'moderate support for H1,' while the abstract claims 'strong evidence,' and the correctness of the headline conclusion depends on this unresolved inferential gap.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies whether cross-cultural digital activism between K-pop fandoms and Black Lives Matter (BLM) discourse on Twitter is driven by belief alignment (H1) or opinion leadership (H2), using the Belief Landscape Framework (BLF) to identify clusters of users with shared expressed beliefs. On a balanced sample of 2,948 users and roughly 29 million tweets from 2020, the authors detect coordinated activity spikes in two attractors (2 and 3) after George Floyd's murder but before BTS's tweet, report that these attractors were among the more heterogeneous mixed-population attractors before the event, and interpret this as support for H1. For H2, they track 272 users who retweeted BTS's tweet and find little movement toward BLM-aligned beliefs, concluding that the tweet amplified but did not initiate activism. They also report a small, marginally significant reduction in between-community correlation as evidence of lasting belief convergence.","tokens_in":19934,"tokens_out":4567,"duration_ms":51859,"significance":"If the central claim were established, it would make a meaningful contribution to theories of cross-cultural digital activism, transnational fandom, and social influence: it would suggest that shared beliefs, rather than celebrity or influencer signaling, drive organic movement spread. The paper also contributes a bilingual application of the Belief Landscape Framework, including a translation pipeline, a sensitivity analysis across half-life settings, and a belief-coherence validation with manual inter-rater agreement. The authors publish code and data in a public repository, which is commendable for reproducibility. However, the headline conclusion rests on a post hoc selection of attractors and an unquantified comparison with non-spiking attractors; the abstract claim of 'strong evidence' is not matched by the results section's 'moderate support for H1'.","major_comments":[{"comment":"The H1 test is post hoc. The authors first detected coordinated spikes in attractors 2 and 3 (Figure 3) and then examined the pre-event homogeneity of those attractors, finding ranks 3 and 5 among 21 attractors in Table 1. Because the attractors were selected on the outcome (the spike), the rank-based observation does not test H1: under the null, some attractors will spike by chance, and any spiking attractor will have some homogeneity rank. The paper acknowledges that the two most heterogeneous attractors (6 and 7) did not exhibit coordinated spikes, but it does not integrate this into a statistical comparison of spiking versus non-spiking attractors. A proper test would define the candidate mixed and BLM-relevant attractors a priori, or use a permutation test that shuffles spike labels across attractors and weeks (preserving marginal spike rates) and compares the realized homogeneity of spiking attractors against the null distribution. As presented, the data are consistent with H1 but do not provide statistically distinguishable support for it.","section":"H1: Belief alignment, Table 1"},{"comment":"The explanation for why the most heterogeneous attractors (6 and 7) did not spike is introduced after observing the outcome: the authors argue that attractor 7 is 'less clearly aligned with BLM discourse' and attractor 6 is a 'high-entropy attractor' defined by fandom content. This is a qualitative, post hoc distinction that is not operationalized as a quantitative predictor. The hypothesis 'belief-aligned mixed attractors spike' is therefore not distinguishable from the alternative 'some attractors spike for unmodeled reasons.' To make the evidence load-bearing, the authors should define an outcome-independent measure of BLM relevance (for example, using the community bias score described in the 'Relative Community Bias' subsection) and test whether spike probability increases with the combination of high mixing and BLM bias. Without such a test, the semantic contrast between attractors 2/3 and 6/7 is descriptive, not confirmatory.","section":"H1: Belief alignment, paragraph contrasting attractors 2/3 with 6/7"},{"comment":"The validity of the entire attractor structure—and hence the H1 and H2 conclusions—depends on the faithfulness of the translated Korean text as input to the English dependency parser. The authors justify the KoBART model with BLEU scores (32.85) and a native-speaker evaluation, but BLEU does not establish that subject-verb-object belief tuples extracted after translation preserve the belief propositions, stance, or emphasis of the original Korean tweets. The belief-coherence validation (71% aligned pairs, Cohen's kappa 0.82) is performed on pairs of translated beliefs within attractors; it does not compare original Korean beliefs with their translated English counterparts. If translation systematically alters belief content, the attractor structure could be an artifact of translation rather than a representation of expressed beliefs. Please report a belief-level translation validation (e.g., back-translation consistency or manual comparison of extracted beliefs from original and translated tweets) to rule out this threat.","section":"Building the Belief Landscape: Translation and belief extraction"},{"comment":"The negative result for H2 is based on the 272 users who retweeted BTS's statement. While the analysis is careful within this group, the broader conclusion that opinion leaders 'do not appear to be a direct cause' of activism is stronger than what this single-leader test can support. The paper acknowledges that exposure to the tweet is not observed and that other potential opinion leaders (e.g., fan 'pillar accounts' discussed in the literature) are not modeled. The observed absence of a belief shift among retweeters is consistent with H2 not operating through BTS's tweet, but it does not rule out opinion leadership by other actors. The abstract should be qualified to refer to the specific test of BTS as an opinion leader.","section":"H2: Opinion Leadership, amplifier analysis"}],"minor_comments":[{"comment":"The abstract states 'strong evidence' for belief alignment, while the results section (H1 summary) states 'we find moderate support for H1.' Please align the language; 'strong evidence' overstates the inferential strength of the post hoc analysis.","section":"Abstract vs. Results"},{"comment":"The claim of a 'statistically significant difference (approximately p < .05)' between the pre-period correlation (r = -0.171) and post-period correlation (r = -0.015) is not supported by a reported test statistic. The 95% confidence intervals listed in Table 2 overlap slightly (Pre upper -0.077, Post lower -0.093), so a proper test (e.g., Fisher r-to-z for independent or dependent correlations) should be described with its p-value.","section":"RQ2: Persistent Belief Change, Table 2"},{"comment":"The homogeneity score uses the absolute value of the difference divided by the sum, so an attractor populated entirely by K-pop users receives the same score (1) as one populated entirely by BLM users. This is appropriate for measuring mixing but is worth stating explicitly so readers do not interpret the score as directional.","section":"Weekly Attractor Homogeneity, Equation (1)"},{"comment":"The caption refers to 'the six attractors exhibiting the most heterogeneity prior to the murder,' but Table 1 lists ten such attractors. Please clarify the selection of the six shown in Figure 3 (e.g., attractors 1, 2, 3, 6, 7, 9, 10 appear in Table 1; the figure's subset is not precisely defined).","section":"Figure 3 caption"},{"comment":"The sentence 'We find moderate support for H1; although coordinated spikes do not appear in the most heterogeneous attractors...' uses 'although' where a conjunction such as 'and' or 'but' would be clearer; the current phrasing suggests the exception is surprising but the logic of the hypothesis is not specified.","section":"Results, H1 summary paragraph"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the scope of a computational social science venue and the case study is topical, but the central claim currently rests on a post hoc selection of attractors. I would be willing to accept a revised version that adds an outcome-independent null-model test for H1 (for example, a permutation test over attractors and weeks), tempers the abstract to 'moderate support,' and includes a belief-level translation validation. If the authors do not add such a test, the paper should be reframed as an exploratory case study rather than a test of H1 versus H2. I saw no concern about citation behavior or duplicate submission; the anonymized repository is appropriate for double-blind review."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper is a careful, transparent case study with a genuine methodological contribution, but the central H1 result is selected post hoc and the abstract oversells it. If you're reading for the method, it's worth your time; if you're reading for the conclusion, treat it as tentative.\n\nWhat's new: they apply Introne's Belief Landscape Framework to a cross-linguistic setting, adding Korean-to-English translation and validating the pipeline with a model-based coherence check (kappa 0.82) and a half-life sensitivity analysis (ARI 0.56-0.84). That's real work. The H2 amplifier analysis—tracking the 272 users who retweeted BTS—is a clever way to get at direct influence, and the null result is interesting even if underpowered.\n\nWhere it's soft: H1 is the load-bearing claim. The authors first find which attractors spike, then check their homogeneity. That's selection on the outcome. They report that attractors 2 and 3 are \"among the most heterogeneous\" but they are not the most; the two most mixed (6 and 7) do not spike, and the explanation for why is qualitative. There's no permutation test, no multiple-testing correction across 21 attractors, and no comparison of spiking vs non-spiking attractors. The paper itself says \"moderate support\" in the results, but the abstract says \"strong evidence.\" That gap should be fixed. The H2 null also lacks confidence intervals or a power analysis, so \"little support\" is not the same as \"evidence against.\" RQ2 is explicitly exploratory, so that's fine.\n\nThe circularity concern is real but not fatal: the framework is the co-author's own, and its validity is not independently re-established. The coherence validation helps, but it does not test whether attractor structure maps onto actual beliefs. Also, the data/code link is anonymized, so I can't verify reproducibility right now, though they claim to share it.\n\nBottom line: worth a serious referee, but the authors should redo the H1 test as an outcome-independent comparison, or at least soften the abstract. If they do, this could be a useful methods paper for studying cross-cultural belief dynamics.\n\nRecommendation: send it out to review, but the referee should press hard on the post hoc selection.","headline":"Solid cross-lingual case study with a real methodological contribution, but the headline belief-alignment result is selected post hoc and the abstract overstates it.","tokens_in":20466,"tokens_out":3088,"would_cite":false,"duration_ms":31728,"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":"K-pop fans' engagement with Black Lives Matter was driven by shared beliefs, not by BTS's endorsement, this study argues.","keywords":["belief alignment","opinion leadership","digital activism","cross-cultural communication","K-pop fandom","Black Lives Matter","Belief Landscape Framework","Twitter analysis"],"falsifier":"A replication that pre-registers which attractors count as 'pre-mixed' before observing any spike, then checks whether the two communities spike in the same attractors during a new surge (for example, another police-killing event or another celebrity statement), would settle H1. For H2, comparing the 272 amplifiers' weighted attractor bias in the 24 hours immediately after BTS's tweet against a matched control group of non-amplifier K-pop users would show whether the tweet itself shifted belief position.","tokens_in":19504,"feed_emoji":"📣","tokens_out":5102,"duration_ms":55237,"temperature":0.7,"pith_summary":"This paper asks why a social movement jumps a cultural and linguistic border: do followers act because influential figures tell them to, or because they already hold beliefs the movement expresses? Using the convergence of K-pop fandom and the Black Lives Matter discourse on Twitter after George Floyd's murder, the authors argue the second explanation is primary. K-pop users spiked inside belief clusters that were already mixed between the two communities before BTS's supportive tweet, and the fans who retweeted the group did not measurably move toward BLM-aligned beliefs afterward. The study concludes that celebrity statements amplify activism rather than initiate it, and reports a small increase in belief similarity between the two communities after the interaction.","feed_headline":"Belief alignment, not idols, drove K-pop fans' BLM surge","feed_subtitle":"Retweeting BTS barely moved fans toward BLM; the surge began in belief spaces they already shared.","key_machinery":"The Belief Landscape Framework (BLF) is the central method: it parses tweets into subject-verb-object belief statements, embeds them with a cross-lingual language model, reduces them to two dimensions with UMAP, finds density peaks ('belief attractors') in user belief-vector space, and tracks which users occupy each attractor over time. The load-bearing pieces are the attractor homogeneity score, which measures how mixed a region's active K-pop and BLM population is in a given week, and the community bias score, which measures whether an attractor's belief content leans BLM or K-pop. The argument uses coordinated activity spikes in pre-mixed, BLM-leaning attractors as evidence for H1, and the stability of amplifier users' weighted bias across the BTS tweet as the test of H2.","core_discovery":"The paper's central claim is that belief alignment—people joining a distant movement because its expressed beliefs resonate with their own—is the primary driver of cross-cultural digital activism, while opinion leadership plays a secondary, amplifying role. The evidence comes from a belief landscape built from roughly 29 million tweets from 2,948 K-pop and BLM users: the only attractors with coordinated spikes from both communities between Floyd's murder and BTS's tweet were attractors 2 and 3, which were among the most heterogeneous (pre-mixed) regions and centred on race, police brutality, and accountability. The 272 K-pop users who retweeted BTS showed almost no shift in their attractor bias across the pre-murder, pre-tweet, and post-tweet periods, and their modest move toward BLM-leaning content happened before the tweet. The paper reads this as fan activism arising from shared beliefs, with the idol's statement acting as validation and amplification rather than the cause. It also finds a weak but statistically significant reduction in negative cross-community correlation between pre- and post-periods, suggesting slight belief convergence.","pith_inferences":["The post-hoc selection of spiking attractors is the point where the paper's own evidence is weaker than its abstract; a pre-registered analysis that names the mixed attractors before observing the spike would be a sharper test of the belief-alignment hypothesis.","If the amplification interpretation generalizes, celebrity donations and statements may function more as resources and signals to already-aligned networks than as persuasion, implying that activism campaigns should invest in grassroots belief infrastructure.","The same pipeline could test whether the pattern holds when a celebrity endorses a foreign movement with no prior history in the fandom; a sudden spike in homogeneous, celebrity-driven attractors in that setting would support a stronger opinion-leadership effect.","The retweet-as-endorsement assumption could be checked by building a follow-up model that uses quote tweets, where users can criticize or argue with the retweeted content, to see whether the 272 amplifiers were agreeing or merely broadcasting."],"forward_implications":["Cross-cultural digital activism is better modelled as belief resonance than as influence cascades; campaigns that hope to cross borders should look for pre-existing shared belief clusters rather than rely on celebrity endorsements alone.","Celebrity or influencer statements can be expected to amplify an already-engaged audience, not to manufacture engagement from disengaged followers.","The method can be applied to other language pairs and movements to map when communities converge around shared moral themes even without shared hashtags or languages.","The slight rise in cross-community belief similarity suggests that even brief co-engagement can push two publics toward common discourse, though the paper treats this effect as exploratory rather than causal."],"supporting_citations":[{"why":"Supplies the Belief Landscape Framework used throughout the analysis, including the attractor construction and the five-week exponential moving average.","marker":"Introne 2023"},{"why":"Provides the prior #MatchAMillion case study showing that ARMY self-organizes through distributed pillar accounts and shared values; the paper builds on this to interpret the amplifiers as hidden opinion leaders.","marker":"Park et al. 2021"},{"why":"Justifies treating retweets as indicative of belief alignment or affiliative endorsement, which is the key proxy for testing the opinion-leadership hypothesis.","marker":"Barberá 2015"},{"why":"Supports the practice of using large-scale retweet patterns as a measure of political stance and social grouping.","marker":"Darwish et al. 2020"},{"why":"Provides one of the two machine translation models evaluated for converting Korean tweets to English before belief parsing.","marker":"Kudugunta et al. 2024"},{"why":"Provides the KoBART translation model actually selected for the cross-linguistic pipeline because of its more consistent and complete Korean-to-English translations.","marker":"SKT-AI 2020"}],"fun_headline_variants":["Belief alignment, not idols, drove K-pop BLM activism","Shared beliefs beat celebrity in K-pop BLM surge","Why K-pop fans joined BLM: beliefs, not BTS","Beliefs, not idols, explain K-pop BLM support","K-pop fans' BLM move rooted in beliefs, not stars"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the belief-landscape clustering of translated, noisy tweets captures users' actual expressed beliefs rather than topic similarity alone; if the spiking attractors look mixed because they share surface topics instead of shared beliefs, the conclusion that belief alignment drove the engagement collapses.","fun_headline_variants_meta":{"raw":{"variants":["Belief alignment, not idols, drove K-pop BLM activism","Shared beliefs beat celebrity in K-pop BLM surge","Why K-pop fans joined BLM: beliefs, not BTS","Beliefs, not idols, explain K-pop BLM support","K-pop fans' BLM move rooted in beliefs, not stars"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000228,"raw_usage":{"total_tokens":1493,"prompt_tokens":981,"completion_tokens":512,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":597,"completion_tokens_details":{"reasoning_tokens":424}},"tokens_in":597,"tokens_out":512,"duration_ms":5763,"temperature":1.0,"reasoning_tokens":424,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:19:35.899100+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A replication that pre-registers which attractors count as 'pre-mixed' before observing any spike, then checks whether the two communities spike in the same attractors during a new surge (for example, another police-killing event or another celebrity statement), would settle H1. For H2, comparing the 272 amplifiers' weighted attractor bias in the 24 hours immediately after BTS's tweet against a matched control group of non-amplifier K-pop users would show whether the tweet itself shifted belief position.","supporting_citations":[],"review_version":1}