{"id":"43a7ed07-79c6-4487-8172-367aca55ef79","arxiv_id":"2508.20152","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":10,"one_line_summary":"A bibliometric study of 4,563 recurrence-plot papers shows exponential growth that slows after 2019, with machine-learning applications as the newest hot topic.","lead":"This paper analyzes a self-curated database of more than 4,500 publications on recurrence plots, a tool for detecting repeated patterns in time-series data. It maps how the field has grown since 1987 and which journals, institutions, and research topics now dominate it.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Growth-rate change points and post-2019 saturation are not formally quantified; database completeness may be time-dependent.","rationale":"The paper's main contribution is a quantitative portrait of the recurrence-plot field's growth and topical evolution. The most load-bearing element is the claimed exponential growth phases and the post-2019 saturation, because that directly supports the conclusion that the field is reaching a limit. However, these growth rates and breakpoints are derived from eyeballed fits in Fig. 3, without formal uncertainty quantification. Additionally, the database is self-curated via citation alerts and snowballing, so its completeness is likely time-dependent: papers that do not cite the seed literature may be missed, and recent publications may be under-captured if they are not yet indexed or cite newer work. The paper acknowledges missing publications but provides no sensitivity analysis linking missingness to the growth-rate estimates. If the missing fraction has increased since 2019, the slowdown is an artifact. This is separate from, though related to, the reader's concern about overall representativeness. The proposed tests—segmented regression with confidence intervals and an independent bibliometric cross-check—would settle whether the quantitative growth claims are robust. Since the reader's CONDITIONAL verdict already calls for formal uncertainty quantification, my concern reinforces that stance without moving the verdict.","tokens_in":17912,"tokens_out":3845,"duration_ms":42018,"concrete_test":"1) Re-fit annual publication counts with segmented regression (e.g., R's segmented package or a Bayesian change-point model) to estimate breakpoints and slopes with confidence intervals. Check whether 2008 and 2019 fall in the credible/confidence regions, and whether the 2019–2024 slope is significantly different from the 2008–2019 slope. 2) Independently query Scopus and Web of Science for 2015–2024 publications with 'recurrence plot' or 'recurrence quantification' in title/abstract/keywords; compare annual counts to the database. If the missing fraction grows after 2019, the saturation claim is suspect. Both tests use the shared Zenodo data and standard bibliometric tools.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—that publication growth followed exponential rates e^{0.221t} (1987–2008), e^{0.095t} (2008–2019), and e^{0.028t} (2019–2024) with a saturation breakpoint—is not supported with formal uncertainty. The rates and change points in Fig. 3 are described qualitatively ('around 2008', 'around 2019') with no confidence intervals, model selection, or sensitivity analysis. This matters because the database is built by citation-alert snowballing from seed papers: completeness may vary systematically with time. If post-2019 papers are under-captured (e.g., because they cite newer literature or appear in venues less covered by alerts), the apparent slowdown would be an artifact. The paper acknowledges missing publications and 9% missing affiliations but does not test how these affect the growth fits. A piecewise exponential fit with unknown breakpoints, and a cross-check against an independent Scopus/WoS search, would determine whether the saturation claim is robust.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript analyzes a custom bibliographic database of 4,563 publications on recurrence plots, recurrence quantification analysis, recurrence networks, and related methods, curated by the author over more than 20 years. It reports: (i) an exponential growth in publications of about e^{0.221t} from 1987 to 2008, slowing to e^{0.095t} from 2008 to 2019 and e^{0.028t} from 2019 to 2024, with a corresponding saturation in the author community; (ii) journal-level publication patterns; (iii) a GMM-based clustering of papers into 21 topical subjects, with cluster 4 (recurrence plots as features for machine learning) identified as the fastest-growing recent topic; (iv) within-community citation statistics, including an approximate e^{-0.07t} decay in citations over time and a list of milestone papers; (v) institutional and author activity rankings; and (vi) country-level and co-author collaboration networks. The paper concludes that the field is lively but approaching saturation, with machine learning as a major future direction.","tokens_in":18225,"tokens_out":4960,"duration_ms":58487,"significance":"If the descriptive claims are robust, the paper provides a valuable quantitative portrait of nearly four decades of a specific methodological field. Its strengths include the long-term curated database, the explicit definition of the in-community citation measure, and the public availability of the data and scripts at Zenodo. The author is also commendably transparent about known gaps (missing publications, 9% missing affiliations, name ambiguity). However, the central quantitative claims about growth-rate change points, saturation, and the temporal evolution of clusters rest on informal fits and a database whose completeness may vary systematically over time. These concerns are not merely cosmetic: the saturation conclusion and the 'fast-growing cluster 4' narrative are load-bearing, and neither is currently supported with formal uncertainty or sensitivity analysis.","major_comments":[{"comment":"The central claim of three growth regimes with breakpoints 'around 2008' and 'around 2019' is based on visual inspection of log-linear plots. No confidence intervals, formal change-point tests, or model-comparison statistics are given for the rates e^{0.221t}, e^{0.095t}, e^{0.028t} or for the community rates e^{0.316t}, e^{0.216t}, e^{0.134t}. Because the saturation conclusion depends entirely on these breakpoints, please fit a piecewise exponential (or Poisson/negative-binomial) model with estimated breakpoints and report parameter uncertainty and model selection. Also test sensitivity to the choice of endpoint years and to excluding 2025 (an incomplete year).","section":"§3.1, Figs. 3 and 5"},{"comment":"The database is constructed by citation alerts from Web of Science and Scopus plus manual curation by the author, who is also one of the most cited authors within the database. Completeness may be time-dependent: recent papers that cite newer literature or appear in venues not covered by the alerts could be under-captured, which would make the post-2019 slowdown an artifact. The paper acknowledges missing publications but does not quantify their effect. Please add an independent validation, e.g., a Scopus/Web of Science keyword search for RP/RQA papers, and sensitivity analyses: with/without papers missing affiliations, with/without the author's own papers, and with/without self-citations. This is essential for the saturation claim and for institutional/author rankings.","section":"§2 and §3.1"},{"comment":"The choice of 21 clusters is only partially supported: the text says the Davies-Bouldin index has a local minimum at 22 but the selected number is 21, and the silhouette score shows a plateau. The temporal analysis applies 'the same clustering schema to a subset' but it is unclear whether this means re-clustering the subset or projecting onto the full-data clusters; if clusters are re-estimated, cluster labels may not be comparable across years. Since the claim that cluster 4 is rapidly growing since 2014 is a key result, please specify the exact clustering pipeline, report cluster stability (e.g., bootstrap or repeated runs), and clarify how cluster identity is maintained over time.","section":"§3.2, Table 3, Fig. 8"},{"comment":"The fits e^{-0.07t} for citation decay, the N^{-1.1} citation distribution, and e^{-0.19t} for author activity are reported without fitting method, goodness-of-fit, or uncertainty. These are secondary descriptive claims, but they should be either given proper statistical support (e.g., least-squares or MLE fits with confidence intervals) or described as approximate visual guides.","section":"§3.3 and §3.5, Figs. 9, 10, 12"}],"minor_comments":[{"comment":"The text refers to 'clusters 18 and 119'; this should presumably be '18 and 19'.","section":"§3.2"},{"comment":"Reconcile the caption and text: the caption says the dashed line indicates the local minimum and the selected cluster number 21, but the text states the local minimum is at 22.","section":"§3.2, Fig. 7"},{"comment":"The sentence about papers 'frequently cited shortly after publication, typically within the same year or shortly thereafter' seems inconsistent with Fig. 9, which shows citations peaking 2–4 years after publication. Please clarify.","section":"§3.3"},{"comment":"The text says 'the top ten papers have more than 200 citations', but Table 4 lists 12 papers with more than 200 citations. Adjust the wording or the table.","section":"Table 4"},{"comment":"The sentence 'Including the current year, we have 38 years of publications (but last year not shown)' is unclear, since Fig. 2 stops at 2024. Please state explicitly which years are plotted and why 2025 is excluded.","section":"§3.1"},{"comment":"Reference [8] contains stray HTML anchor text; formatting, not content, appears corrupted.","section":"Reference list"}],"recommendation":"major_revision","confidential_remarks":"The paper is a single-author bibliometric self-study by the curator of the database, and the author is also a highly cited actor within the field. This is not disqualifying, but it makes the formal uncertainty and sensitivity analyses I request in the major comments especially important. The data and scripts are available, so the requested analyses are feasible within the manuscript's scope. If the authors provide the formal change-point analysis and an independent completeness check, the paper could be publishable; without them, the core saturation/narrative claims remain unsupported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nShort version: this is a capable, honestly written bibliometric portrait of the recurrence-plot field, built on a database the author has maintained for over twenty years. It is the first comprehensive analysis of this literature, and it ships the data and scripts. The main caveat is that the headline quantitative claims—the exponential growth rates and the change points around 2008 and 2019—are supported by eyeballed fits rather than formal breakpoint estimation. That doesn't sink the paper, but it does cap what we can conclude from the growth curves.\n\nWhat's genuinely new: the curated database itself (4,563 papers, with references, affiliations, topics) is a real contribution, and the analysis draws a useful map of the field: 21 topical clusters, the emergence of machine-learning applications since 2014, the citation landscape, and the collaboration network. The paper is refreshingly candid about its limitations: missing publications, 9% missing affiliations, Chinese-name ambiguity, in-community citations only. It doesn't overclaim.\n\nSoft spots, in proportion: the growth-rate change points are chosen from log-linear plots by eye, with no confidence intervals or model selection. The GMM cluster count (21) rests on the Davies-Bouldin index but also on \"seems quite appropriate\"—subjective, though not scandalously so. More subtly, the database is built from citation alerts around seed papers, so completeness could be time-dependent; a cross-check against an independent Scopus/WoS search would strengthen the post-2019 saturation claim. The closure issue (author is the database curator and a dominant author within it) is real, but the institution and citation rankings are presented descriptively, not as an evaluation, and the author flags the problem.\n\nVerdict: this deserves a serious referee. The descriptive portrait is useful to newcomers, funders, and anyone writing the history of nonlinear time series analysis. It won't change practice, but it does its job. I'd recommend acceptance with relatively light revision (formal change-point analysis or at least a sensitivity check; a sentence or two on time-dependent completeness). It may not belong in a top general journal, but as a field-specific bibliometric study it is solid and reproducible.","headline":"A solid, honest bibliometric portrait of the recurrence-plot field built on a long-maintained curated database; the headline growth-rate change points are eyeballed rather than formally tested, but the paper ships its data and is worth a serious referee.","tokens_in":18674,"tokens_out":1868,"would_cite":false,"duration_ms":20852,"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":"This paper claims that recurrence-plot research expanded exponentially for two decades, then slowed sharply after 2008 and nearly plateaued after 2019, with machine-learning applications the only fast-growing niche.","keywords":["recurrence plots","recurrence quantification analysis","bibliometrics","publication growth","co-author networks","topic clustering","citation analysis","machine learning"],"falsifier":"An independent comprehensive search of the published literature using a broad but literal query for 'recurrence plot' or equivalent method names in title and abstract, without relying on the author's curated bibliography, would produce annual publication counts for 2019-2024 much higher than the reported ~3% growth; if the independent growth is instead much steeper, the claimed saturation is an artifact of database curation.","tokens_in":17813,"feed_emoji":"📉","tokens_out":8954,"duration_ms":96220,"temperature":0.7,"pith_summary":"The paper sets out to give a quantitative life history of the research field built around recurrence plots, a two-dimensional visual and numerical tool for detecting when a dynamical system returns to past states. It analyzes a bibliography of 4,563 relevant publications that the author has maintained for more than twenty years. The central result is that the field's annual output grew nearly exponentially from 1987 to 2008, then slowed markedly after 2008 and again after 2019, approaching saturation; meanwhile, a cluster of papers using recurrence plots as features for machine learning has been the only rapidly growing niche. The paper also maps the 21 topical subjects, the most-cited works, active institutions, and collaborative communities in the field.","feed_headline":"Publication growth in recurrence science slowed to a crawl","feed_subtitle":"From 4,563 papers: growth fell from ~25% to ~3% a year; machine learning is the only fast-growing niche.","key_machinery":"The load-bearing object is the author-maintained BibTeX bibliography (4,563 analysed entries, monthly updated for over two decades) and the analysis pipeline around it: exponential fits to yearly counts; lemmatization plus Gaussian-mixture clustering (a statistical grouping method) of titles, abstracts, and keywords, with the Davies-Bouldin index and silhouette score fixing the cluster number; in-community citation counts from two reference sources; and country-level and co-author networks partitioned by a community-detection algorithm. The recurrence plot itself—a binary matrix that marks when a trajectory returns close to an earlier state—is the shared object that all clusters and mileston","core_discovery":"The paper's central claim is that the recurrence-plot field has followed a specific life cycle visible in its own publication record. From a curated bibliography of 4,563 papers, the author finds annual output grew roughly exponentially from 1987 to 2008 at about e^{0.221t}, then slowed to e^{0.095t} (2008-2019), then e^{0.028t} (after 2019), approaching saturation. Gaussian-mixture clustering of titles, abstracts, and keywords yields 21 topical subjects; the one with rapid recent growth is cluster 4, where recurrence plots serve as image-like features for machine-learning classifiers. In-community citations are highly concentrated: 43% of papers are never cited by other database papers, whi","pith_inferences":["The post-2019 'saturation' may partly be an artifact of the curator's monthly update being unable to track an increasingly diffuse literature; an independent broad query on 'recurrence plot' in titles and abstracts would test whether true output has also plateaued.","Machine-learning papers that use recurrence plots as input images probably cite the broader ML literature more than the RP/RQA canon, so in-community citations understate their impact; full citation counts would likely show cluster 4 growing even faster than the curve here.","Author-name ambiguity (especially for Chinese names) is a measurable confound; re-running the author-level and co-author-network analysis on author-identifier-verified identities would quantify how much it shapes community sizes.","The same change-point-fitting recipe could be applied inside cluster 4 to see whether deep-learning-driven recurrence research is itself heading toward saturation or still in its exponential phase."],"forward_implications":["If the measured rates hold, the field is no longer in rapid self-driven expansion; further growth must largely come from new application communities.","Cluster 4's growth since 2014 signals that recurrence plots are now material for deep-learning pipelines, not only nonlinear-dynamics diagnostics.","The heavy citation concentration implies that a small set of canonical texts anchors the community and that most contributions integrate the method into applications rather than extend it.","The 2008 and 2019 rate breaks give future bibliometric updates clear tests: maintaining e^{0.028t} confirms a plateau; an upturn would indicate a new wave.","The database's in-community citation counts are only a lower bound on real influence, since external citations are not counted."],"supporting_citations":[{"why":"Defines the recurrence plot, the object whose literature the whole database and growth analysis track.","marker":"[1]"},{"why":"The most-cited review in the field and a hub for the 2007 citation peak.","marker":"[2]"},{"why":"One of the foundational papers introducing recurrence quantification analysis.","marker":"[4]"},{"why":"The companion foundational RQA paper; a major early citation peak.","marker":"[5]"},{"why":"Introduced vertical-line RQA measures, a milestone in the 2002 citation peak.","marker":"[6]"},{"why":"Launched the recurrence-network concept that forms a distinct topical cluster.","marker":"[7]"},{"why":"Established recurrence networks as a paradigm and a 2010 citation milestone.","marker":"[8]"},{"why":"The curated bibliography is the database analyzed in the paper.","marker":"[30]"},{"why":"Davies-Bouldin index used to choose the number of topical clusters.","marker":"[47]"},{"why":"Silhouette score used to validate the 21-cluster subject partition.","marker":"[48]"}],"fun_headline_variants":["Recurrence plot research hits growth plateau","Recurrence plots: field's boom fades, ML niche thrives","From 25% to 3% growth: recurrence science matures","Recurrence plot field slows, machine learning surges","Recurrence science growth collapses; only ML climbs"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The load-bearing premise is that the author-curated database is complete and accurate enough to represent the entire recurrence-plot literature; the paper itself acknowledges likely missing publications, 9% missing affiliations, and unresolved author-name ambiguity, especially among Chinese names.","fun_headline_variants_meta":{"raw":{"variants":["Recurrence plot research hits growth plateau","Recurrence plots: field's boom fades, ML niche thrives","From 25% to 3% growth: recurrence science matures","Recurrence plot field slows, machine learning surges","Recurrence science growth collapses; only ML climbs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00015,"raw_usage":{"total_tokens":962,"prompt_tokens":601,"completion_tokens":361,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":345,"completion_tokens_details":{"reasoning_tokens":279}},"tokens_in":345,"tokens_out":361,"duration_ms":4320,"temperature":1.0,"reasoning_tokens":279,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T15:13:25.076966+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"An independent comprehensive search of the published literature using a broad but literal query for 'recurrence plot' or equivalent method names in title and abstract, without relying on the author's curated bibliography, would produce annual publication counts for 2019-2024 much higher than the reported ~3% growth; if the independent growth is instead much steeper, the claimed saturation is an artifact of database curation.","supporting_citations":[{"cited_title":"Eckmann, S","cited_arxiv_id":null,"evidence_quote":"Defines the recurrence plot, the object whose literature the whole database and growth analysis track."},{"cited_title":"Marwan, M","cited_arxiv_id":null,"evidence_quote":"The most-cited review in the field and a hub for the 2007 citation peak."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"One of the foundational papers introducing recurrence quantification analysis."},{"cited_title":"Marwan, N","cited_arxiv_id":null,"evidence_quote":"Introduced vertical-line RQA measures, a milestone in the 2002 citation peak."},{"cited_title":"Marwan, J","cited_arxiv_id":null,"evidence_quote":"Launched the recurrence-network concept that forms a distinct topical cluster."},{"cited_title":"http://www","cited_arxiv_id":null,"evidence_quote":"The curated bibliography is the database analyzed in the paper."}],"review_version":1}