A comprehensive scientometric investigation mapping a decade of global data science scholarship has revealed a dramatic surge in publication volumes alongside persistent geographic imbalances and evolving thematic boundaries. Analyzing twenty-seven thousand one hundred and eight peer-reviewed documents indexed in the Scopus database between 2015 and 2025, the study delineates the structural transformation of data science from an emergent computational crossroad into an established research discipline. The underlying dataset encompassed contributions distributed across nine thousand and five distinct publishing sources, drawing input from seventy-six thousand three hundred and thirty-six unique authors. Over the ten-year evaluation window, scholarly output expanded substantially, driven by strategic national investments in computational infrastructure and automated predictive technologies. However, despite the worldwide relevance of large-scale predictive modeling and computational analysis, collaborative networks remain distinctly regionalized. Only twenty-two point five nine percent of the investigated publications involved international co-authorship, demonstrating that cross-border research integration continues to lag behind the rapid domestic expansion observed across leading research economies.
Methodologically, the investigation employed established scientometric tools, specifically the Biblioshiny package within Bibliometrix and VOSviewer, to track productivity patterns, institutional affiliations, keyword co-occurrences, and international co-authorship structures. To ensure methodological rigor and systematic filtering, the authors implemented the Preferred Reporting Items for Systematic Reviews and Meta-Analyses, or PRISMA, workflow across identification, screening, eligibility, and inclusion phases. The initial database query across titles, abstracts, and author keywords yielded twenty-nine thousand two hundred and eighty-three raw records published in English. After restricting the dataset to finalized publications and eliminating incomplete entries, twenty-seven thousand one hundred and eight items remained for formal evaluation. The documentary corpus exhibited a balanced division across communication channels, led by ten thousand eight hundred and forty-seven conference publications and ten thousand four hundred and thirty journal articles. In addition, the collection contained one thousand nine hundred and twelve book chapters, one thousand seven hundred and sixty-six review articles, and nine hundred and one complete books, underscoring the hybrid disciplinary heritage of data science.
Longitudinal growth dynamics were quantified through the mathematical formulations of Relative Growth Rate and Doubling Time. The Relative Growth Rate, designated as the logarithmic increase in accumulated publications per unit of calendar time, was paired with Doubling Time to characterize expansion velocity. Setting 2015 as the baseline starting point with an assigned growth rate of zero, researchers noted that publication output accelerated sharply following 2018. This inflection coincided with major state-level artificial intelligence policy frameworks, including China’s Next Generation Artificial Intelligence Development Plan and the United States National Artificial Intelligence Initiative. Across the evaluated timeframe, the calculated Relative Growth Rate attained its peak value of one point one eight in the final recorded year, while Doubling Time fell to its lowest point of zero point five nine years. Conversely, the longest Doubling Time was recorded in 2016 at three point three one years, demonstrating how the field progressively condensed the time necessary to double its cumulative scientific literature.
To evaluate individual productivity structures, the authors tested Lotka’s Law, an empirical bibliometric formulation establishing that the number of scholars producing a specific number of articles is inversely proportional to that productivity squared. In the observed corpus, sixty-one thousand four hundred and sixty-two researchers, representing eighty point five two percent of all identified contributors, published exactly one paper. Only eleven point three three percent of authors contributed two documents, and the proportion continued to fall steeply at higher publication frequencies. While author productivity broadly followed the inverse power distribution predicted by the classical model, mathematical comparisons revealed consistent negative deviations, with the largest divergence occurring among two-paper contributors. This skewed pattern reflects the presence of a compact productivity elite comprising authors with eighteen or more published papers, surrounded by an extensive periphery of transient contributors who engaged in data science through isolated interdisciplinary projects across healthcare, engineering, and social science domains.
Citation impact analysis highlighted influential individual contributors and revealed clear structural ties between academic institutions and commercial corporate laboratories. Zhang, Xiao from Meta in the United States emerged as the most cited author within the corpus, accumulating fourteen thousand six hundred and twenty-six citations. This total was heavily driven by a foundational 2019 Nature Biotechnology paper on reproducible and extensible microbiome data science utilizing the QIIME 2 platform. Wang, Jiaqun from Carnegie Mellon University and Liu, Tang from Nanyang Technological University in Singapore secured prominent citation totals through collaborative treatises on federated learning published in Foundations and Trends in Machine Learning, which garnered more than seven thousand citations. Impactful contributions also originated from Wang, Xiao at Beihang University, Zhang, Kai at the University of Connecticut, Li, Xiang at Massachusetts General Hospital, and Chen, Ying at IBM Research-Almaden, underscoring the vital contributions of clinical medicine and corporate technology laboratories in driving high-impact computational research.
The study examined source distribution using Bradford’s Law of Scattering, which sorts publishing venues into concentric zones of decreasing productivity containing equal numbers of published articles. The twenty-seven thousand one hundred and eight documents were mapped into three zones: Zone 1 comprised one hundred and three nucleus journals producing eight thousand nine hundred and sixty-eight papers; Zone 2 included one thousand five hundred and one journals producing nine thousand one hundred and ninety-seven papers; and Zone 3 contained seven thousand four hundred and one journals containing eight thousand nine hundred and forty-three papers. With an empirical Bradford multiplier of nine point seven five, theoretical calculations projected twenty-one thousand eight hundred and seventy-four journals in the third zone, contrasting sharply with the actual eight thousand nine hundred and forty-three titles. The resulting error margin of twenty-one point zero seven percent demonstrated that data science resists tidy theoretical scattering, dispersing across diverse venues led by conference proceedings while specialized outlets like Patterns establish disciplinary independence.
Thematic mapping based on keyword co-occurrence positioned research concepts on two-dimensional coordinates defined by network density, measuring internal cohesion, and network centrality, measuring external connectedness. The upper-right quadrant highlighted motor themes with high centrality and density, led by data science with six thousand seven hundred and fifty-nine occurrences, machine learning with three thousand nine hundred and eighty-seven, and big data with one thousand six hundred and ninety-three occurrences. These core drivers were joined by natural language processing, Python, education, statistics, artificial intelligence, social media, and COVID-19 analytics, reflecting rapid methodological response to the global pandemic. The basic themes quadrant gathered fundamental, highly central concepts that remain less internally integrated, including deep learning, data mining, data analytics, classification, data visualization, and neural networks. Conversely, niche topics such as specialized data science applications in education demonstrated strong internal coherence but marginal centrality, while emerging clusters like data and data science occupied exploratory positions with low centrality and low density.
A three-field Sankey plot linked prolific authors, high-volume publication outlets, and dominant descriptors, illustrating how intellectual authority has consolidated through repetitive focus on computational architectures. Leading contributors such as Zhang, Yilong, Chen, Ying, and Li, Jiaxin maintained dense ties to high-capacity publication venues, most notably the ACM International Conference Proceeding Series and the Journal of Physics: Conference Series. In the overarching keyword co-occurrence network, the primary search term data science exhibited the highest betweenness centrality at thirty-eight point three seven nine and the leading PageRank value of zero point zero nine five, confirming its role as an integrating hub between machine learning, artificial intelligence, and big data. However, the network structure also revealed pronounced technical clustering, whereas socially oriented applications, algorithmic ethics, and public policy topics occupied peripheral nodes, indicating that technical refinement continues to outpace socially embedded inquiry across mainstream funding channels.
Geographic production metrics established the United States and China as the primary centers of global data science scholarship, followed closely by India and the United Kingdom. In author appearances across the corpus, the United States registered thirty-three thousand four hundred and three occurrences associated with one hundred and eight thousand eight hundred and twenty-five citations, while China logged ten thousand seven hundred and eighty appearances and twenty-three thousand four hundred and twenty-three citations. India recorded ten thousand three hundred and eighty-six author appearances and ten thousand six hundred and ninety-one citations, followed by the United Kingdom with six thousand five hundred and fifty-four appearances and twenty thousand seven hundred and seventy-seven citations. Co-authorship analysis confirmed the United States as the dominant structural node, contributing eight thousand six hundred and seventy-seven collaborative documents, one hundred and ninety-nine thousand and sixty-one citations, and a total link strength of four thousand nine hundred and ninety-one. Meanwhile, dense intra-European ties linked Germany, the United Kingdom, France, and Italy, contrasting with lower collaboration volumes across African and Latin American institutions.
The authors emphasized several methodological and structural limitations that contextualize their empirical findings. The exclusive reliance on the Scopus database omitted non-indexed journals, regional repositories, and non-English publications, creating an institutional bias toward well-resourced Western and Asian publishing infrastructures while underrepresenting Global South scholarship. In addition, the zero growth rate recorded for 2015 represented an artifact of selecting that year as the baseline origin, while publication figures for 2025 reflected a partial collection window captured on July 15, 2025. Because citation accrual is inherently time-dependent, recently published papers from 2023 to 2025 and rapidly expanding national systems such as China exhibit lower average citations compared to older Western contributions. Furthermore, bibliometric indicators capture documented publication outputs rather than direct causal drivers such as private corporate research funding or geopolitical policy strategies, requiring cautious interpretation when assessing global research dynamics.
Subject of Research: Bibliometric analysis of global data science research publications from 2015 to 2025
Article Title: A decade of data science research: insights from a bibliometric study
Article References: Verma, R., Kumar, A., Pabreja, K., & Kumar, A. (2025). A decade of data science research: insights from a bibliometric study. Discover Data, 3(1), Article 61. https://doi.org/10.1007/s44248-025-00097-5
Image Credits: AI Generated
DOI: 10.1007/s44248-025-00097-5
Keywords: data science, bibliometrics, scientometrics, machine learning, artificial intelligence, Lotka’s law, Bradford’s law, research collaboration, decade, data, science, research
Source: bioengineer.org
