Purpose: The rapid expansion of digital technologies and the increasing availability of large-scale datasets have positioned data mining as one of the most influential research domains. Given its interdisciplinary nature, understanding the co-word structure and thematic development of data-mining research has become increasingly important. Although bibliometric studies have explored international trends using databases such as Scopus and Web of Science, relatively little attention has been devoted to Persian-language research outputs. This study investigates the conceptual structure and thematic evolution of Persian data-mining research indexed in the Islamic World Science Citation Center (ISC) over a twenty-year period (2005–2024). Specifically, it identifies research themes, examines their temporal evolution, maps conceptual relationships among keywords, and determines the strategic positions of themes within the field.
Methodology: This descriptive study has an applied approach and uses bibliometric analysis tools. Data were manually collected from ISC in April 2026 using different variations of the Persian equivalent of the term “data mining.” The initial search retrieved 595 records. After expert review and screening to exclude conceptually irrelevant publications, 589 articles published between 2005 and 2024 were retained for analysis. Because ISC does not provide comprehensive bibliographic exports comparable to those of international databases, data were extracted manually using spreadsheet software. Author keywords were identified and standardized through expert-led data cleaning. Synonyms, singular and plural forms, spelling inconsistencies, and spacing and orthographic variations were merged into unified terms to improve analytical accuracy. The general keyword “data mining” was excluded because of its ubiquitous presence and limited discriminatory value. Biblioshiny was used to generate thematic evolution maps, co-occurrence networks, and strategic diagrams. These analyses enabled the identification of dominant themes, conceptual clusters, emerging topics, and patterns of thematic continuity and transformation over time. They also provided a basis for examining how keyword relationships and theme positions changed across the selected periods.
Findings: The 589 articles contained 1,734 keywords, averaging approximately three keywords per publication. Decision trees (n = 78) and artificial neural networks (n = 74) were the most frequent keywords, followed by clustering (n = 53), prediction (n = 43), and support vector machines (n = 35). Temporal analysis showed the continued presence of established methodological keywords, alongside increased occurrences of “machine learning” and “random forest” in later periods. Because “machine learning” is a broad paradigm, its increased frequency alone does not demonstrate the adoption of a specific advanced algorithm. Thematic evolution analysis showed continuity and branching of selected keywords across the three periods, including links between methodological terms and application domains. “Decision trees” and “clustering” persisted across periods, while some methodological and application-related terms appeared or became more frequent later. For example, thematic flows connected decision trees with support vector machines and artificial neural networks, while clustering was linked with application-related terms such as breast cancer. These patterns describe the observed continuity and branching of keywords and should not be interpreted as evidence of a general transition toward more advanced methods or as a definitive developmental trajectory for the entire field.
Co-word network analysis identified ten conceptual clusters. The network included both methodological keywords and application-related terms, illustrating the coexistence of data-mining methods and the domains in which those methods were used. The clusters therefore capture both algorithmic approaches and their reported application contexts. The interpretation is limited to Persian-language articles indexed in ISC and reflects the co-occurrence structure observed in the selected dataset. Strategic diagram analysis showed that the identified clusters were distributed mainly among the basic, specialized, and emerging/declining quadrants, with no cluster located in the motor-theme quadrant. This distribution should be interpreted cautiously because the network was constructed from keywords occurring at least three times and therefore represents the structure of the available dataset rather than the thematic maturity of data-mining research in Iran as a whole.
Conclusion: The findings suggest a gradual transformation in Persian data-mining research, from an initial focus on foundational algorithms to the application of data-mining techniques in specialized domains. Nevertheless, the absence of mature motor themes and the continued prominence of basic topics indicate that strong, independent thematic trajectories have yet to emerge in the analyzed dataset. The results also suggest that Persian-language research has largely adapted internationally established paradigms, rather than generating distinctive theoretical contributions. Strengthening interdisciplinary collaboration, expanding engagement with emerging research areas, and enhancing connections with international scientific networks may facilitate further thematic development, maturity, and innovation. These findings describe the keyword and co-occurrence patterns observed in the selected Persian-language ISC corpus and should not be generalized to the entire data-mining research output of Iran. Future studies using broader databases, larger keyword networks, and complementary bibliometric analyses could provide a more comprehensive assessment of thematic development and support comparison with international research trends.