Submitted:
30 June 2026
Posted:
01 July 2026
You are already at the latest version
Abstract
This study introduces a proof-of-concept methodology for hypergraph-based decomposition aimed at identifying coherent thematic structures within OpenAlex bibliometric records on science governance spanning the period from 2022 to 2026. A total of 17,034 records were exported, containing 153,765 keywords, of which 8,153 were unique. The keyword sets of these bibliometric records were conceptualized as hyperedges within a hypergraph. The partitioning of the hypergraph into four balanced blocks was executed using the Mt-KaHyPar algorithm, applying the developer-recommended parameters for high-quality partitioning. The underlying pipeline operations proceed by first applying balanced hypergraph partitioning to extract tightly coupled blocks, and then filtering bibliometric records by three keywords from blocks. To articulate the specific themes of the key terms associated with each block, several highly relevant publications and a concise summary of the potential research topic were generated for each block usingPerplexity.ai. The results demonstrate the rationality of utilizing a co-occurrence criterion based on three or more keywords within a single bibliometric record. Two of the four extracted blocks contain the most frequently occurring terms; one reflects the theme of corporate governance, politics, and sustainability, while the other encompasses business, political science, finance, sociology, and law. Conversely, the remaining two blocks tend to reflect emergent domains that may hold significant future research potential: one centers on landscape assessment, landscape design, and soil water, while the other focuses on digital marketing, tax reform, and customer intelligence. The suggested workflow can serve as a methodological template for mapping science and identifying current research topics.
Keywords:
science governance
; bibliometric analysis
; OpenAlex records
; hypergraph partitioning
; Mt-KaHyPar program
Introduction
The contemporary domain of science government is confronted with a fundamental challenge: the exponential proliferation of scientific output, the increasing fragmentation of knowledge, and the progressive erosion of conventional disciplinary boundaries [1]. These interrelated phenomena impede the effective coordination of transformative interdisciplinary research, particularly in contexts demanding rapid responses to global environmental, biological, and technological exigencies [2,3].
The pertinence of the present inquiry is underscored by the pressing need for analytical instruments capable of both mapping and forecasting the trajectory of scientific development. However, extant approaches to textual analysis remain largely confined to dyadic relationships among individual indicators, despite the fact that key scholarly dimensions—including keyword co-occurrence, co-authorship networks, and citation patterns—are inherently characterized by multivariate attribute sets rather than binary associations [4].
Algorithmically, graph models are commonly employed for such tasks due to their interpretability; however, in thematic analysis, they tend to oversimplify the underlying data structure by reducing multilogical, context-dependent relationships to binary entities co-occurrences. This entails a loss of information regarding the simultaneous presence of multiple terms within a single research topic. In practice, this limitation is particularly evident in VOSviewer—now a de facto standard for bibliometric visualization—whose maps effectively delineate dense clusters and central terms, yet whose graph-based formalism conflates co-occurrence frequency with semantic relatedness. Consequently, more intricate yet crucial structures—such as multicomponent problem formulations, and latent interdisciplinary linkages—may be partially obscured or distorted, thereby constraining the analytical depth of conclusions drawn in the context of identifying relevant research agendas.
A potential remedy lies in replacing graph-based representations with hypergraph models. In this framework, hyperedges are formed from keyword sets extracted from scientific publications, enabling the preservation of multidimensional thematic relationships—unlike conventional graphs, which are restricted to pairwise connections [5]. Hypergraph partitioning thus yields not merely frequent term pairs, but stable multi-word contexts and thematic clusters that correspond to coherent research domains [6]. This is particularly valuable for literature reviews and science mapping, as the resulting components encapsulate key concept sets that delineate specific problem areas, thereby facilitating interpretation of thematic structure and identification of emergent or underdeveloped directions. Furthermore, formal partitioning provides quantitative indicators—such as intra-block hyperedge density and component overlap—suitable for comparative assessment of research activity and novelty. When coupled with automated keyword extraction, hypergraph decomposition supports large-scale, reproducible construction of problem ontologies across extensive corpora, enhancing both the objectivity and efficiency of identifying salient research questions for strategic planning and science policy.
For partitioning keyword sets into balanced blocks, mt-KaHyPar [7,8] is recommended, as it is specifically designed for high-quality graph and hypergraph partitioning under balancing constraints while minimizing edge cuts. This property is particularly relevant to thematic analysis: balanced partitions yield clusters of comparable size, facilitating cross-comparison of research areas, detection of thematic gaps, and construction of problem maps.
However, its application in scientific practice remains limited. Existing literature predominantly addresses algorithmic performance and computational characteristics, with little attention to keyword-based decomposition of scholarly texts. Thus, while the method is methodologically robust and promising, its practical adoption for this purpose is not yet widespread [9,10].
The proposed hypergraph-based decomposition approach is not merely a technical refinement of existing bibliometric techniques; it rests upon a substantive theoretical foundation that underscores the epistemic primacy of triadic structures over dyadic ones. Across semiotics, computational linguistics, network science, and cognitive research, a consistent principle emerges: stable meaning and coherent thematic context arise not from pairwise associations, but from triadic configurations. That is, the hyperedge should contain at least three terms to explain the subject matter of the texts.
In classical semiotics, Peirce established that signification is irreducibly triadic: meaning emerges only through the interplay of Object, Representamen, and Interpretant, whereas dyadic relations remain mechanically associative and semantically unstable. This insight was formalized in linguistics by Ogden and Richards through their Semantic Triangle—Word, Concept, Thing—demonstrating that a stable conceptual frame requires three vertices. In knowledge engineering, Sowa’s conceptual graph theory similarly posits the Subject–Predicate–Object triplet as the minimal unit capable of expressing a complete proposition and resolving ambiguity in automated text processing; recent AI research further reinforces this by modeling knowledge as a triadic entity comprising Index, Meaning, and Mode. Network analysis corroborates this view through the principle of triadic closure: in semantic word networks, triads form dense, stable clusters with high clustering coefficients, whereas pairwise connections tend to be transient and noise-prone (Easley & Kleinberg, 2010). Cognitive studies likewise confirm that human comprehension shifts from associative perception to genuine contextual understanding precisely when processing triadic rather than dyadic information.
In classical semiotics, the Peircean framework establishes that signification is an irreducibly triadic phenomenon, wherein meaning emerges exclusively through the reciprocal interplay of the object, representamen, and interpretant [11]. Conversely, dyadic relations are constrained to mechanical association, rendering them semantically unstable. This conceptual architecture was subsequently formalized within linguistics by Ogden and Richards via the Semantic Triangle (comprising word, concept, and thing), which demonstrates that a stable conceptual frame necessitates three interconnected vertices [12,13].
In knowledge engineering, Sowa’s conceptual graph theory similarly posits the subject–predicate–object triplet as the foundational minimal unit capable of expressing complete propositions and mitigating ambiguity in automated text processing [14]. Recent study introduces TAXAL (Triadic Alignment for eXplainability in Agentic LLMs), a triadic framework integrating cognitive, functional, and causal dimensions of explainability. TAXAL offers a unified, role-sensitive scaffold for the design, evaluation, and deployment of explanations in diverse sociotechnical contexts. [15].
Network science corroborates these structural principles through the mechanism of triadic closure; within semantic word networks, triads consolidate into dense, topologically stable clusters characterized by high clustering coefficients, whereas pairwise configurations remain highly susceptible to transient noise [16]. This structural stability is mirrored in cognitive processing, where empirical evidence confirms that human comprehension transitions from baseline associative perception to genuine contextual understanding precisely when processing triadic rather than dyadic informational architectures [17].
This theoretical consensus directly informs the methodological rationale of the present study. The hypergraph partitioning procedure, implemented via mt-KaHyPar, yields balanced blocks of keywords; crucially, the emergence of a block containing exactly three co-occurring keywords—a triadic node set—is not a coincidental artifact. Rather, it corresponds to a minimally closed semantic system, a stable thematic nucleus that preserves multidimensional contextual relationships lost in conventional graph-based pairwise models. Such three-term configurations simultaneously anchor a coherent research topic, filter out spurious associations, and render visible the latent interdisciplinary linkages that dyadic approaches tend to obscure. Thus, the presence of three-node keyword blocks within a partitioned hypergraph provides both a theoretically grounded indicator of thematic integrity and a practically operationalizable unit for constructing meaningful, noise-resilient science maps—thereby enhancing the validity and interpretability of large-scale bibliometric analysis for research planning and science policy.
For generating hyperedges from keywords, it is rational to use OpenAlex [18,19], which is a compelling source of bibliometric data due to its comprehensive coverage, open access, and continuous updates, which together eliminate many of the limitations inherent in proprietary databases. Of particular methodological significance is its collaboration with Leiden University’s Centre for Science and Technology Studies (CWTS), which has resulted in the integration of a controlled-vocabulary keyword field derived from the CWTS in-house classification system. As demonstrated by recent empirical validations [20], this unified multi-level hierarchical infrastructure successfully bridges citation-network clustering with large-scale semantic text analysis, thereby enabling highly robust global overlay mapping across diverse academic fields without the constraints of traditional journal-level categorization
The aim of this study
- To identify consistent thematic structures in OpenAlex bibliometric records related to science management.
- To propose a research pipeline that includes preprocessing bibliometric records exported from OpenAlex, constructing a hypergraph based on data from the “Keywords,” the partitioning of the hypergraph into balanced thematic blocks, and then the filtering of bibliometric records based on sets of three keywords from each block in order to formulate specific topics based on key terms associated with each block.
Materials and Methods
The bibliometric records of contemporary research in the field of Science Governance spanning the period from 2022 to 2026 were sourced from the OpenAlex global aggregator and exported in JSON format.
The OpenAlex database was queried using a search string filtered by publication year and document type (article). Pagination and request offsetting were managed via a cursor. A typical API query is presented below:
The merging of JSON files, the extraction of specific fields—namely, “id”, “doi”, “title”, “publication_year”, “fwci”, “citation_percentile”, “keywords”, and “concepts”—and the subsequent generation of the final CSV file were performed using the jq — command-line JSON processor (available at https://github.com/jqlang/jq).
The annual CSV files were merged into a single consolidated dataset using xan, a command-line tool designed for processing CSV files (available at https://github.com/medialab/xan). The final dataset comprised 17,034 records. The “keywords” field contained a total of 153,765 terms, of which 8,153 were unique. Only two records within this field were empty. In this study, the data from the “keywords” field were utilized to construct the hyperedges of a hypergraph.
Prior to constructing the hypergraph in .hgr format, a data preprocessing step was performed on the keywords. Specifically, all terms were converted to lowercase to ensure case insensitivity, and spaces within multi-word expressions were replaced with underscores. This normalization process was executed to facilitate string manipulation and ensure that multi-word terms were treated as single semantic units during subsequent analysis.
To generate the hypergraph.hgr file, a text-to-ID mapping procedure was implemented, wherein each unique term from the “keywords” field was assigned a distinct, sequential integer identifier to construct an unweighted hypergraph.
The partitioning of the resulting hypergraph into four balanced blocks was executed using Mt-KaHyPar (Multi-Threaded Karlsruhe Hypergraph Partitioner). The partitioning process was configured to minimize the km-1 metric under a strict balance constraint. Specifically, the execution was performed via the command-line interface with the following parameters:
Mt-KaHyPar -h hyperedges.hgr -k 4 -e 0.03 -o km1 --preset-type highest_quality -w > OUT_Mt-KaHyPar.txt
Here, the hypergraph was split into blocks with an allowed imbalance tolerance of (3%). The optimization objective was set to the km-1 metric (-o km1), which directly minimizes the connectivity of hyperedges across different partitions. To ensure maximum partition accuracy, the --preset-type highest_quality configuration was applied. The -w flag (--write-partition-file) was enabled to output the computed vertex-to-block assignments into a dedicated partition file, while the standard text output of the execution log was redirected to OUT_Mt-KaHyPar.txt
Following the hypergraph partitioning, a reverse-mapping procedure was executed to restore the normalized keyword strings from their respective integer identifiers. Given that each resulting block contained approximately 2,000 keywords, identifying the dominant thematic orientation of each partition directly from the raw term list was unfeasible. To address this, a keyword ranking and filtering approach was implemented based on term co-occurrence frequencies within the original “keywords” records. The primary filtering criterion required at least three keywords from the same block to co-occur within a single publication’s “keywords” field. As established in the Introduction, a co-occurrence threshold of three keywords provides a robust thematic representation while effectively narrowing the analytical sample. From a graph-theoretic perspective, this procedure is equivalent to extracting induced sub hypergraphs by selecting only those hyperedges that contain at least three nodes belonging to the target block.
The extraction of relevant records based on their keyword content was implemented using ugrep, a high-performance file pattern searcher available at https://github.com/Genivia/ugrep.
Within the subset of records assigned to each block, the keyword combinations exhibiting the highest co-occurrence frequencies were identified. These high-frequency term patterns were subsequently utilized as query vectors to retrieve the corresponding publications. To ensure the selection of high-impact research, the retrieved documents were filtered, retaining only those with the highest values in the “citation_percentile” field; these specific publications were operationalized as the core literature accurately representing the thematic focus of the respective block.
Furthermore, to complement the empirical data, the same subsets of prominent keywords were used as prompts within the Perplexity.ai generative engine to synthesize concise thematic summaries.
The English translation of the text was performed using Google Gemini and DeepL Translator, followed by thorough manual editing to ensure accuracy and stylistic consistency.
Results and Discussion
To evaluate the structural characteristics of the analyzed dataset, it is necessary to outline the core parameters of the search query executed in OpenAlex—an open-access bibliometric platform that has gained significant academic validity through its recent integration with the CWTS Leiden University initiative. The primary metadata collection was completed on June 9, 2026, using a targeted query within the article titles and abstracts (search.title_and_abstract=science+governance). Restricting the publication window to 2022–2026 and filtering by document type (type:article) yielded a baseline sample of 17,034 publications.
For context, as of June 26, 2026, the total volume of works indexed by OpenAlex exceeded 317.8 million entries. A comparative analysis of syntax processing revealed that the database treats the “+” symbol as a logical AND operator; an explicit query for science+and+governance returned an identical result of 18,365 articles, reflecting a natural dataset growth between June 9 and June 26, 2026.
The methodological choice to restrict the search scope to titles and abstracts (search.title_and_abstract) is critical for ensuring thematic relevance. For comparison, a broader global search (search=science+governance) for the 2023–2026 period returned 519,541 documents, which introduces excessive noise. Conversely, a strict exact-match phrase query (“science+governance”) within titles and abstracts for 2022–2026 yielded only 116 articles. Thus, using the unquoted keywords science and governance within titles and abstracts provides an optimal balance, capturing a broader yet thematic context for analyzing science governance literature.
According to the OpenAlex analytical framework, the dataset retrieved via the primary search query is predominantly categorized into the five thematic domains outlined in Table 1.
These baseline metrics serve as a valuable benchmark for comparison with the empirical findings generated in the course of this study.
Next, we examine the baseline characteristics of the keywords data field within the analyzed records.
Core Characteristics of the Keywords Field
The final dataset of 17,034 records contains a total of 153,765 terms, yielding an average of approximately 9.0 keywords per publication. This high density indicates that OpenAlex assigns terms from a standardized, controlled vocabulary rather than relying solely on the traditional 5–7 author-defined keywords. Consequently, this structural feature makes the dataset more comparable to indexed keyword systems, which significantly enhances metadata depth.
The frequency distribution of these keywords across the analyzed records reflects the global thematic orientation of the sample. The most prominent terms are summarized in Table 2.
Note: For formatting convenience, the underscores in compound terms in Table 2, Table 3, Table 4, Table 5 and Table 6 have been replaced with spaces.
The results presented in Table X are significant, as they establish the baseline keywords that will subsequently be aggregated into distinct blocks.
Keyword Hypergraph Partitioning into Four Blocks
According to the Mt-KaHyPar execution logs, the hypergraph partitioning produced the following optimization metrics: (primary objective function), , , and an imbalance factor of . The distribution of nodes (keywords) across the designated blocks is as follows: ; ; ; and . Notably, block 0 contains the minimum number of nodes, whereas blocks 2 and 3 contain the maximum.
As justified in the Introduction and Methods sections, a co-occurrence of at least three terms is considered optimal for identifying the thematic focus of a text. Consequently, after mapping the node IDs back to normalized keywords, we filtered the records to isolate those containing at least three terms from the respective block within their Keywords field. The frequency of a keyword within these filtered records was utilized as a proxy for its significance.
The subsequent tables present the top 30 keywords ranked by frequency for each block.
In Table 2, the term landscape_design appears 7 times out of its 10 total occurrences across the entire dataset, indicating its strong dominance within the thematic scope of Block 0. To transition from this keyword-based overview to specific literature, a systematic article selection workflow was applied to each block.
First, to filter the core vocabulary, a frequency threshold was optimized for each block to yield approximately 20 key terms. For Block 0, a threshold of 5 occurrences isolated 17 terms, whereas a threshold of 4 expanded the list to 30; thus, a frequency of 5 was selected as an optimal compromise to avoid overextending the manuscript.
Next, we extracted the subset of records where the Keywords field contained these filtered terms. To identify representative publications from this subset, documents were ranked by the co-occurrence frequency of the selected keyword combinations. In cases with multiple matching publications, the highest-ranking articles were selected based on their citation_percentile metric, and their DOIs were retrieved for full-text analysis.
Block 0 is characterized by two frequent term clusters in the keyword field: landscape_assessment, landscape_planning, landscape_design and biomass_(ecology), soil_science, soil_water.
Representative publications:
- Article [21] Terms in abstract: landscape restoration activities ... and governance aspects; natural and social sciences ... highly complex territories.
- Article [22] Terms in abstract: landscape sustainability science (LSS); landscape management, governance.
- Article [23] Terms in abstract: Science Data Bank; governance of future terrestrial ecosystems.
Perplexity.ai (https://perplexity.ai/) offers the following thematic summary:
“This topic studies how landscape design and planning can support resilient, subsistence-based agricultural systems with sound soil–water management, while considering ecological biomass and human health outcomes (e.g., nutrition via food systems affecting gut microbiome), analyzed through cohort and informal-sector perspectives.”
The table encompasses not only digital marketing terminology—such as digital_marketing, marketing_research, tax_reform, marketing_management, and customer_intelligence—but also concepts from mathematics and computer science, including algebra_over_a_field, support_vector_machine, attractor, random_forest, domain_analysis, generator_(circuit_theory), cloud_computing_security, software_construction, and quantum_computer. Thus, it reflects both methodological approaches and their domains of application.
Filtering for terms with frequency ≥8 and retrieving the most relevant publications yields the following list for digital_marketing, marketing_research, and marketing_management.
Representative publications:
- Article [24]. Terms in abstract: development of business marketing science; marketing management governance.
- Article [25]. Terms in abstract: design-science methodology; importance of governance and responsible implementation.
- Article [26]. Terms in abstract: science and technology studies (STS); outlining governance implications.
- Article [27]. Terms in abstract: psychology, sociology, organizational studies, and political science; organizational development, governance, and social reform.
Perplexity.ai (https://perplexity.ai/) provides the following brief topic characterization (keywords with weights):
- Short summary: this topic links tax-policy questions (reform, credits, indirect taxes) with applied analytics and modeling tools from marketing and dynamical systems, alongside considerations of moral cognition affecting compliance or consumer behavior. The mixture suggests an interdisciplinary study (policy + data science + behavioral ethics).
This block presents the most frequent terms, three of which effectively define the core thematic focus of the collected bibliometric records: corporate_governance, politics, and sustainability. Notably, the set also includes terms pertaining to topic analysis and scientometric research—namely, thematic_analysis, Scopus, systematic_review, bibliometrics, and even citizen_journalism. This combination suggests that, within the given context, bibliometrics is oriented more toward identifying emerging research themes—as indicated by thematic_analysis, Scopus, systematic_review, and citizen_journalism—than toward computing formal performance metrics.
Block 2, comprising 19 terms with frequency ≥500, constitutes a substantially large segment, as evidenced by file size: block2Top.txt → 479 KB, compared to block12Top.txt → 5 KB and block0Top.txt → 3 KB.
The most frequent three-term co-occurrence is corporate_governance, climate_change, politics.
Representative publications:
- Article [28]. Terms in abstract: applying climate science to adjudicate legal disputes; climate governance role.
- Article [29]. Terms in abstract: climate change from a “lens” in the social sciences; non-democratic governance arrangements.
- Article [30]. Terms in abstract: public understanding of science; risk governance and economic pathways.
This block may be regarded as a core cluster for subsequent research.
Perplexity.ai (https://perplexity.ai/) provides the following brief topic characterization (keywords with weights):
- Short summary: an interdisciplinary field studying how corporate governance and political-institutional forces shape corporate responses to sustainability and climate risks, assessed through accountability/transparency frameworks and qualitative methods, with sectoral applications (e.g., health care) and bibliometric interest.
The terms related to Block 3 listed in Table 5 rank second in frequency after the terms related to Block 2.
The foundational terms in this table are political_science, business, sociology, economics, and law. The pervasive role of digitalization in addressing contemporary challenges is evidenced by the presence of computer_science and artificial_intelligence. Governance-related concepts also feature prominently, represented by management, environmental_resource_management, knowledge_management, public_administration, and process_management.
The most frequent three-term co-occurrences are political_science–sociology–law and business–political_science–finance. Political_science serves as the common node; for sociology, the significant associate is law, whereas for business it is finance.
For the first triplet, the publications with the highest normalized citation scores are:
For the second triplet:
- Article [34] – score: 0.9990. Terms in abstract: Web of Science; decentralized governance systems.
- Article [35] – score: 0.9987. Terms in abstract: political science; Governance by Numbers 2.0.
- Article [36] – score: 0.9987. Terms in abstract: management sciences and financial accounting; corporate governance.
Perplexity.ai (https://perplexity.ai/) provides the following thematic summary:
“This topic studies how political, legal, economic, organizational, and technical systems interact to govern environmental and social outcomes—especially when technology, data, and private actors intersect with public-interest goals like sustainability, health, and equitable resource management.”
The final two tables collectively indicate the overarching dominance of the corporate_governance + political_science theme. This is unsurprising, as science policy and corporate governance priorities substantially shape research funding allocation and, consequently, the prevailing research agendas.
Synthesis of Thematic Blocks
The four blocks collectively reveal a coherent interdisciplinary landscape organized around governance as the central integrating theme, albeit operationalized through distinct disciplinary lenses and levels of analysis.
The topics of the publications related to the most popular terms in each block can be summarized as follows.
Block 0 (landscape assessment, landscape planning, landscape design; biomass, soil science, soil water) grounds the abstract governance frameworks in tangible spatial and ecological contexts. This cluster addresses landscape sustainability, restoration, and terrestrial ecosystem governance, integrating natural and social sciences to support resilient agricultural systems and sound resource management.
Block 1 (digital marketing, marketing research, marketing management) centers on applied analytics and decision systems in business contexts, with emerging attention to moral cognition, governance, and responsible implementation of AI-driven tools. Its orientation is primarily organizational and behavioral, bridging data science with marketing strategy and policy compliance.
Block 2 (corporate governance, climate change, politics) shifts the focus to macro-level institutional forces, examining how corporate governance and political frameworks shape organizational responses to sustainability and climate risks. This cluster is characterized by strong emphasis on accountability, transparency, and qualitative assessment methodologies, with clear bibliometric and thematic-analysis overtones.
Block 3 (political science, business, sociology, economics, law) provides the foundational disciplinary backbone, highlighting the interplay between political, legal, economic, and organizational systems in governing environmental and social outcomes. The presence of computer science and AI underscores the digitalization imperative, while management-related terms reflect attention to process and resource governance. The dominance of the corporate_governance–political_science nexus here points to the recursive relationship between science policy, funding priorities, and research agendas.
Cross-cutting synthesis: Across all blocks, governance emerges as the overarching conceptual umbrella—whether at the organizational (Block 1), institutional-political (Block 2), disciplinary-theoretical (Block 3), or spatial-ecological (Block 0) level. Digitalization and data-driven methodologies permeate each cluster, reflecting a broader epistemic shift toward computational and analytics-based approaches. The recurring presence of terms such as systematic_review, bibliometrics, and thematic_analysis across blocks also indicates a strong reflexive dimension: the dataset itself is concerned with mapping and evaluating research landscapes. Finally, the explicit or implicit role of policy—from tax reform to climate governance to landscape restoration—suggests that these bibliometric patterns capture an applied, problem-driven research ecosystem where scientific knowledge is increasingly expected to inform regulatory and strategic decision-making.
In sum, the four blocks depict a research environment that is methodologically data-intensive, thematically governance-centric, and substantively oriented toward addressing complex socio-environmental challenges through interdisciplinary, policy-relevant inquiry. Comparing the obtained results with the data presented in Table 1 indicates that they are more specific and closer to the content of the analyzed bibliometric records.
Conclusion
A proof-of-concept research pipeline was proposed and tested. It includes preprocessing bibliometric records exported from OpenAlex, constructing a hypergraph based on “Keywords” data, partitioning the hypergraph into balanced thematic blocks, and filtering bibliometric records based on sets of three keywords from each block to formulate specific topics based on the key terms associated with each block and, based on these, selecting publications that relevantly disclose the topics of the blocks.
Based on the proposed approach to analyzing bibliometric records exported from OpenAlex, four topical blocks were identified. Each block is characterized by a specific set of keywords reflecting the following thematic areas:
Based on the proposed approach to analyzing bibliometric data exported from OpenAlex, four thematic blocks were identified. Each block is characterized by a specific set of keywords and a selected publications reflecting the following subject areas:
Block 0 grounds abstract governance frameworks in spatial and ecological contexts. It integrates natural and social sciences to address landscape sustainability, terrestrial ecosystem governance, and resilient resource management.
Block 1 focuses on applied analytics and decision systems within business. It bridges data science with marketing strategy, emphasizing organizational behavior, moral cognition, and the responsible implementation of AI.
Block 2 examines macro-level institutional forces, specifically how corporate governance and political frameworks shape responses to climate risks. Its emphasis lies in accountability, transparency, and qualitative assessment methodologies.
Block 3 provides the foundational multidisciplinary backbone, linking political, legal, economic, and technological systems. The prominence of the corporate governance–political science nexus highlights the recursive relationship between science policy, digital transformation, and research agendas.
Future research should extend this approach by comparing the partitioning of keyword-constructed hypergraphs into balanced blocks using alternative algorithms alongside Mt-KaHyPar. Special emphasis should also be placed on science funding as a vital component of broader science governance. Practically, the proposed workflow offers a systematic way to process large-scale bibliometric data and map interconnected thematic areas.
Funding
The work was funded by the Ministry of Science and Higher Education of the Russian Federation (State Assignment No. 125021302095-2).
References
- Linder F, Spear J, Nowotny H, Scott P, Gibbons M. Re-thinking science: knowledge and the public in an age of uncertainty. Contemporary Sociology. 2003;32(2):255. [CrossRef]
- Jordana J, Holesch A, Schmitt L, et al. Institutional transformations of global governance – key challenges for international organisations. irpp. 2024;6(3):463-481. [CrossRef]
- Lawrence MG, Williams S, Nanz P, Renn O. Characteristics, potentials, and challenges of transdisciplinary research. One Earth. 2022;5(1):44-61. [CrossRef]
- Lung RI, Gaskó N, Suciu MA. A hypergraph model for representing scientific output. Scientometrics. 2018;117(3):1361-1379. [CrossRef]
- Zhang L, Guo J, Wang J, Wang J, Li S, Zhang C. Hypergraph and uncertain hypergraph representation learning theory and methods. Mathematics. 2022;10(11):1921. [CrossRef]
- Zhang H, Liu X, Zhang J. Hegel: hypergraph transformer for long document summarization. In: Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics; 2022:10167-10176. [CrossRef]
- Gottesbüren L, Heuer T, Maas N, Sanders P, Schlag S. Scalable high-quality hypergraph partitioning. ACM Trans Algorithms. 2024;20(1):1-54. [CrossRef]
- Çatalyürek Ü, Devine K, Faraj M, et al. More recent advances in (Hyper)graph partitioning. ACM Comput Surv. 2023;55(12):1-38. [CrossRef]
- Li Y, Zhang X, Bai X, Bai S, Jiang Z. Predicting co-word links via heterogeneous graph convolutional networks. Sci Rep. 2025;15(1):23143. [CrossRef]
- Chigarev B. Hypergraph partitioning of openalex keywords to reveal thematic blocks in the economics of science. Preprint posted online June 9, 2026. [CrossRef]
- Friedman A, Thellefsen M. The Peircean theory of AI: Advancing text generation through Peirce’s triadic model, speculative grammar, and methodeutics. DASC. 2025;8:50-70. [CrossRef]
- McElvenny J. Ogden and richards’ the meaning of meaning and early analytic philosophy. Language Sciences. 2014;41:212-221. [CrossRef]
- Zhao Y. The meaning of the meaning of meaning. In: Philosophical Semiotics. Springer Nature Singapore; 2022:51-67. [CrossRef]
- Sowa J. Conceptual graphs. Knowledge-Based Systems. 1992;5(3):171-172. [CrossRef]
- Herrera-Poyatos D, Peláez-González C, Zuheros C, Tejedor V, Montes R, Herrera F. TAXAL framework: Triadic fusion of cognitive, functional, and causal dimensions for explainability in agentic LLMs. Information Fusion. 2026;134:104389. [CrossRef]
- Stewart IA, Buehler MJ. Higher-order knowledge representations for agentic scientific reasoning. arXiv. Preprint posted online 2026. [CrossRef]
- Isbilen ES, McCauley SM, Christiansen MH. Individual differences in artificial and natural language statistical learning. Cognition. 2022;225:105123. [CrossRef]
- Thelwall M, Jiang X. Is OpenAlex suitable for research quality evaluation and which citation indicator is best? Asso for Info Science & Tech. 2025;76(12):1660-1681. [CrossRef]
- Culbert JH, Hobert A, Jahn N, et al. Reference coverage analysis of openalex compared to web of science and scopus. Scientometrics. 2025;130(4):2475-2492. [CrossRef]
- Haunschild R, Bornmann L. The use of OpenAlex to produce meaningful bibliometric global overlay maps of science on the individual, institutional, and national levels. Konys A, ed. PLoS ONE. 2024;19(12):e0308041. [CrossRef]
- Barrera-Causil C, González-Montañez J. Harmonization approach to spatial and social techniques to define landscape restoration areas in a colombian andes complex landscape. Forests. 2023;14(9):1913. [CrossRef]
- Qiu J, Nassauer JI, Ahern J, et al. Advancing landscape sustainability science: key challenges and strategies for integration with landscape design and planning. Landsc Ecol. 2025;40(2):25, s10980-024-02042-02044. [CrossRef]
- Sun Q, Zhang P, Jiao X, et al. A global estimate of monthly vegetation and soil fractions from spatiotemporally adaptive spectral mixture analysis during 2001–2022. Earth Syst Sci Data. 2024;16(3):1333-1351. [CrossRef]
- Rudianto R, Misrofingah M, Prayoga D, Juliana J, Al Sukri S, Pong KS. Understanding management marketing in digitalization and automation times. EKU. 2022;17(2):110-121. [CrossRef]
- Nguyen TPL. The AIMx framework: integrating marketing mix modeling, attribution, and AI-driven analytics for adaptive decision systems. Futur Bus J. 2026;12(1):121. [CrossRef]
- Li H. Synthetic moral agents or sophisticated mimics? Published online June 30, 2026. [CrossRef]
- Prof. Dr. Yoesoep Edhie Rachmad PD. Moral disengagement theory. Published online 2026. [CrossRef]
- Kotzé LJ, Mayer B, van Asselt H, et al. Courts, climate litigation and the evolution of earth system law. Global Policy. 2024;15(1):5-22. [CrossRef]
- Gabehart KM, Nam A, Weible CM. Lessons from the Advocacy Coalition Framework for climate change policy and politics. Clim Action. 2022;1(1):13. [CrossRef]
- Zhang K, Qu B, Huang Q. Negotiating climate change: Science, policy, and the invisible power embedded in public discourse in Chinese social media. Sammut G, ed. PLoS One. 2026;21(5):e0348708. [CrossRef]
- Shapin S. Hard science, soft science: A political history of a disciplinary array. Hist Sci. 2022;60(3):287-328. [CrossRef]
- Soares L, Cockle KL, Ruelas Inzunza E, et al. Neotropical ornithology: Reckoning with historical assumptions, removing systemic barriers, and reimagining the future. Ornithological Applications. 2023;125(1):duac046. [CrossRef]
- Tassadiq F, Silver J, Kallianos Y, Guma PK. The unending corridor: Critical approaches to the politics, logics and socio-technics of urban corridorisation. Urban Studies. 2025;62(10):1961-1984. [CrossRef]
- Balcerzak AP, Nica E, Rogalska E, Poliak M, Klieštik T, Sabie OM. Blockchain technology and smart contracts in decentralized governance systems. Administrative Sciences. 2022;12(3):96. [CrossRef]
- Steiner-Khamsi G, Martens K, Ydesen C. Governance by numbers 2.0: policy brokerage as an instrument of global governance in the era of information overload. Comparative Education. 2024;60(4):537-554. [CrossRef]
- Affes W, Jarboui A. The impact of corporate governance on financial performance: a cross-sector study. Int J Discl Gov. 2023;20(4):374-394. [CrossRef]
Table 1.
Top five OpenAlex thematic domains within the study dataset.
| Thematic Domain / Topic | Document Count |
|---|---|
| Ethics and Social Impacts of AI | 644 |
| Artificial Intelligence in Healthcare and Education | 447 |
| Research Data Management Practices | 333 |
| Sustainability and Climate Change Governance | 295 |
| Coastal and Marine Management | 280 |
Table 2.
Top 30 most frequent keywords within the study dataset.
| Terms | count | Terms | count | Terms | count |
|---|---|---|---|---|---|
| corporate governance | 7120 | context_(archaeology) | 1452 | medicine | 825 |
| political science | 4366 | engineering | 1322 | climate change | 810 |
| business | 3200 | public relations | 1093 | environmental science | 786 |
| computer science | 2780 | government_(linguistics) | 1062 | field_(mathematics) | 742 |
| sociology | 2330 | sustainable development | 982 | health care | 721 |
| politics | 2185 | management | 944 | knowledge management | 713 |
| economics | 2103 | finance | 895 | work (physics) | 708 |
| law | 1864 | psychology | 891 | environmental planning | 676 |
| sustainability | 1840 | ecology | 875 | social science | 669 |
| geography | 1547 | environmental resource management | 857 | process (computing) | 642 |
Note: Terms are presented in their standardized lowercase, underscore-delimited format as processed to ensure cross-reference consistency with the raw replication dataset.
Table 3.
Top 30 terms in block 0 based on a minimum three-term co-occurrence in the keywords field.
| Terms | Count | Terms | Count | Terms | Count |
|---|---|---|---|---|---|
| cohort | 7 | neighbourhood (mathematics) | 5 | evapotranspiration | 4 |
| landscape design | 7 | homo sapiens | 5 | water use | 4 |
| survival of the fittest | 7 | stem cell | 5 | causal model | 4 |
| subsistence agriculture | 7 | biomass (ecology) | 5 | protectionism | 4 |
| soil water | 6 | national laboratory | 5 | medical ethics | 4 |
| soil science | 6 | gut microbiome | 5 | deception | 4 |
| landscape assessment | 6 | gut flora | 5 | living systems | 4 |
| informal sector | 5 | surface runoff | 4 | causal chain | 4 |
| landscape planning | 5 | cohort study | 4 | epoch (astronomy) | 4 |
| transplantation | 5 | tracing | 4 | cosmic cancer database | 4 |
Table 4.
Top 30 terms in block 1 based on a minimum three-term co-occurrence in the keywords field.
| Terms | Count | Terms | Count | Terms | Count |
|---|---|---|---|---|---|
| tax reform | 16 | attractor | 8 | security information and event management | 7 |
| json | 12 | observability | 8 | random forest | 7 |
| algebra over a field | 11 | observable | 8 | support vector machine | 7 |
| marketing research | 10 | social cognitive theory of morality | 8 | a priori and a posteriori | 7 |
| digital marketing | 9 | moral disengagement | 8 | binary number | 7 |
| tax credit | 9 | moral reasoning | 8 | propulsion | 7 |
| dynamical systems theory | 9 | domain analysis | 8 | information security management | 6 |
| generator (circuit theory) | 9 | operator (biology) | 8 | cloud computing security | 6 |
| marketing management | 8 | indirect tax | 8 | software construction | 6 |
| customer intelligence | 8 | customer advocacy | 7 | quantum computer | 6 |
Table 5.
Top 30 terms in block 2 based on a minimum three-term co-occurrence in the keywords field.
| Terms | Count | Terms | Count | Terms | Count |
|---|---|---|---|---|---|
| corporate governance | 6752 | process (computing) | 628 | democracy | 485 |
| politics | 2046 | thematic analysis | 622 | systematic review | 480 |
| sustainability | 1795 | accountability | 609 | big data | 478 |
| context_(archaeology) | 1416 | stakeholder | 589 | transformative learning | 476 |
| government (linguistics) | 1031 | Scopus | 579 | public health | 469 |
| sustainable development | 952 | transparency (behavior) | 554 | citizen journalism | 455 |
| climate change | 767 | conceptual framework | 538 | China | 443 |
| field (mathematics) | 718 | quality_(philosophy) | 536 | resilience (materials science) | 414 |
| work (physics) | 688 | Key (lock) | 507 | perspective (graphical) | 412 |
| health care | 678 | state (computer science) | 486 | bibliometrics | 405 |
Table 6.
Top 30 terms in block 3 based on a minimum three-term co-occurrence in the keywords field.
| Terms | Count | Terms | Count | Terms | Count |
|---|---|---|---|---|---|
| political science | 4264 | finance | 885 | philosophy | 589 |
| business | 3119 | ecology | 874 | public administration | 584 |
| computer science | 2661 | psychology | 865 | data science | 568 |
| sociology | 2310 | environmental resource management | 857 | engineering ethics | 523 |
| economics | 2096 | medicine | 801 | accounting | 446 |
| law | 1850 | environmental science | 772 | epistemology | 422 |
| geography | 1515 | knowledge management | 712 | economic growth | 412 |
| engineering | 1317 | environmental planning | 673 | marketing | 378 |
| public relations | 1091 | social science | 668 | artificial intelligence | 364 |
| management | 944 | biology | 600 | process management | 348 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.
