Source-linked AI summary
A Global Map of Science Based on the ISI Subject Categories
Loet Leydesdorff, Ismael Rafols
TL;DR
The paper asks whether ISI subject categories can support comprehensive maps of science despite imperfect journal-level classifications. It aggregates citation data by category and applies factor analysis, finding a usable fourteen-factor disciplinary structure while retaining finer-grained limitations.
Problem
The paper addresses whether ISI subject categories can provide a comprehensive science map despite their imperfect match with citation-derived journal structures.
Method
The study aggregates journal-journal citations into citing and cited subject-category matrices and applies exploratory factor analysis at multiple aggregation levels.
Results
Fourteen factors explain 51.8% of variance in the citing projection and 47.9% in the cited projection, with 89% of categories assigned to the same factor in both.
Takeaways & Limitations
ISI subject categories function moderately well for classification and support comprehensive maps at the category level, while journal citations retain finer-grained detail.
Takeaways & Limitations
The resulting map is comprehensible but not conclusive because ISI categories do not match the fine-grained structure derived from aggregated journal citations.
Abstract
from arXiv · showhide
The ISI subject categories classify journals included in the Science Citation Index (SCI). The aggregated journal-journal citation matrix contained in the Journal Citation Reports can be aggregated on the basis of these categories. This leads to an asymmetrical transaction matrix (citing versus cited) which is much more densely populated than the underlying matrix at the journal level. Exploratory factor analysis leads us to opt for a fourteen-factor solution. This solution can easily be interpreted as the disciplinary structure of science. The nested maps of science (corresponding to 14 factors, 172 categories, and 6,164 journals) are brought online at http://www.leydesdorff.net/map06/index.htm. An analysis of interdisciplinary relations is pursued at three levels of aggregation using the newly added ISI subject category of "Nanoscience & nanotechnology". The journal level provides the finer grained perspective. Errors in the attribution of journals to the ISI subject categories are averaged out so that the factor analysis can reveal the main structures. The mapping of science can, therefore, be comprehensive at the level of ISI subject categories.
ISI Subject Categories
The paper examines ISI subject categories as a basis for mapping science, while recognizing that their classifications do not fully match citation-derived structures. Aggregating categories can nevertheless reveal a robust disciplinary organization.
- ISI assigns journals to subject categories using criteria including journal titles and citation patterns.
- The category system cannot be treated as a literary-warrant classification because it cannot be revised retrospectively from the perspective of hindsight.
- The 172 subject categories are finer-grained and more suitable for science mapping than the 22 broad fields.
- The category frequencies and coverage are uneven, ranging from 262 journals in one category to five in another.
- ISI categories match citation-derived journal clusters closely in only somewhat more than 50% of cases, leaving many journals heterogeneous.
- Fourteen factors organize the categories into biology-medicine, physics-materials-engineering-computing, and environment-ecology-geosciences clusters.
Methods
The study constructs separate cited and citing category matrices from 2006 Journal Citation Reports data, normalizes them, visualizes their relations, and determines the factor count empirically.
- The dataset contains 172 citing categories and 175 cited categories derived from the 2006 Journal Citation Reports.
- The cited and citing matrices are normalized with Salton’s cosine, visualized in Pajek, and analyzed by factor analysis in SPSS.
- The number of factors is assessed empirically using eigenvalues, screeplot inspection, and testing against the data.
Results
Factor analysis supports a fourteen-factor disciplinary structure in the citing and cited dimensions, with most subject categories assigned consistently across them. The resulting maps show nested disciplinary and interdisciplinary relations, including routes connecting medical and hard-science domains.
- Matrix structure: 11,577 of 30,100 category-pair cells are zero, making the category matrix denser than the underlying journal-level citation matrix.The zero cells represent 38.46% of the matrix; factor analysis normalizes using Pearson correlations, whereas visualizations use cosine values.
- Factor solution: 14 factors are supported by the screeplot, and extracting more than 14 substantially reduces factor quality.The fourteen-factor solution explains 51.8% of citing-projection variance and 47.9% of cited-projection variance.
- Disciplinary structure: The fourteen factors can be interpreted as disciplines including Physics, Chemistry, Clinical Medicine, Neurosciences, Engineering, and Ecology.The largest factor is Biomedical Sciences, combining core biological sciences, biological methodologies, and medicine-related fields such as Oncology and Pathology.
- Factor solution: 154 of 172 subject categories, or 89%, fall in the same factor in both citing and cited projections.The remaining 18 categories are classified differently across the two dimensions.
- Maps: The maps represent 171 related subject categories at cosine > 0.2, with category colors indicating the fourteen factors and node sizes based on logarithmic citation counts.The disciplinary and subdisciplinary maps are interactively linked, and visual inspection should be supplemented by projections from multiple angles.
- Interdisciplinary structure: Chemistry brokers relations between Physics and Materials Sciences and core Biomedical Sciences, while computer and biomedical categories appear structurally central.Three routes connect the medical and hard-science poles: computer science links to medical specialties, Chemistry brokers between poles, and Engineering and Materials Sciences connect through environmental and biological fields.
Interdisciplinarity
Interdisciplinarity is examined across subject categories, factors, and journals, using betweenness centrality after cosine normalization. The nanoscience case shows that journal-level maps reveal finer interdisciplinary structure than category-level maps.
- Subject categories: Only 1.16% of cited-dimension betweenness centrality is contributed by Nanoscience & nanotechnology, compared with 1.5% in the citing dimension.The cited and citing dimensions therefore yield different positions for the same category.
- Subject categories: Betweenness centrality identifies categories that connect otherwise distinct regions of the science system.Computer science and general biomedical categories appear especially important at the database level.
- Subject categories: The nanoscience category neighborhood contains 37 related subject categories, whose disciplinary affiliations include materials sciences, chemistry, engineering, physics, and biomedical sciences.The category-level environment is organized through factor-colored disciplinary groupings.
- Interpretation: The category map is comprehensible but not conclusive because category boundaries behave like a fuzzy classification.Materials science, biomaterials leads at 5.6% despite weak connection to the core, while Nanoscience & nanotechnology contributes 5.4%.
- Journals: At the journal level, Microelectronic Engineering has the highest betweenness centrality at 47.6%, while journals with “nano” in their titles form a separate, less central cluster.Among 32 nanoscience journals, 21 enter the displayed network at cosine > 0.2; the underlying citation matrix density is 34.2%.
- Journals: Nano Letters shifted from an orientation toward nano-journals to an intermediating position between nano-journals and physical-chemistry journals.This local structural change is visible through journal-level betweenness centrality but not through ISI subject categories.
Conclusions and discussion
The nanoscience analysis finds that ISI subject categories function moderately well for broad classification, while journal citations provide a finer-grained but harder-to-aggregate picture. Factor analysis reduces lower-level variation and supports comprehensive maps.
- Conclusions and discussion: ISI subject categories function moderately well across journal, category, and disciplinary-factor levels, but do not match the fine-grained journal-citation picture.Journal-level aggregation is more detailed, whereas higher-level bottom-up mapping requires assumptions.
- Conclusions and discussion: The trade-off is between fine-grained journal citation maps and comprehensive higher-level maps based on aggregated subject categories.The former are difficult to construct bottom-up at specialty and disciplinary levels without assumptions.
- Conclusions and discussion: Factor analysis statistically averages stochastic variation in journal-to-category assignments and reduces the complexity of the citation data.The resulting maps are reported to match previously published mappings by Boyack, Börner, and Klavans.
Appendix I
Appendix I presents factor-analysis results for 172 ISI Subject Categories in the citing projection.
- Appendix I: The appendix is headed “Factor analysis of 172 ISI Subject Categories” for the citing projection.It identifies the appendix as a factor-loading display rather than a narrative analysis.
- Appendix I: The displayed appendix material reports numerical factor-analysis entries for subject categories in the citing projection.The supplied excerpt provides the table heading but no complete interpretation of its factors.
- Appendix I: Appendix I therefore supplies the quantitative category-by-factor material underlying the citing-side analysis.Its role is to document the factor-analysis results for the 172 categories.
CITING FACTORS
The citing-factor appendix lists factor loadings for ISI subject categories, showing disciplinary concentrations alongside categories with substantial loadings on multiple factors.
- Biology and medicine: Biological and biomedical categories show strong citing-factor loadings, including cell biology at 0.890 and biophysics at 0.818.Other examples include microbiology at 0.622 and physiology at 0.601 on prominent listed factors.
- Cross-factor loadings: Several categories have notable loadings across multiple factors rather than a single dominant loading.Examples include hematology, biology, oceanography, composites, and applications.
- Neurosciences: Neurosciences, behavioral sciences, neuroimaging, and clinical neurology have strong loadings on a common listed factor.Their reported values are 0.870, 0.804, 0.792, and 0.752, respectively.
- Chemistry and physics: Chemistry, physical and crystallography have substantial loadings on the listed chemistry-related factors.Their reported values include 0.622 for chemistry, physical and 0.600 for crystallography.
- Engineering and technology: Engineering, mechanics, mechanical, aerospace, and related categories display strong loadings on another listed factor.Examples include mechanics at 0.831, mechanical at 0.757, and aerospace at 0.453.
INFECTIOUS
The infectious-science entries report component coefficients for tropical medicine, parasitology, mycology, veterinary sciences, and environmental science.
- Tropical medicine has coefficients 0.565 and 0.366 across the reported components.
- Parasitology has coefficients 0.300, 0.111, and 0.560 across the reported components.
- Mycology has coefficients 0.419, 0.127, 0.461, and 0.432 across the reported components.
- Veterinary sciences has coefficients 0.151, -0.103, and 0.429 across the reported components.
- Environmental science has coefficients 0.136, 0.176, 0.114, and 0.762 across the reported components.
ENVIRONMENTAL SCIENCES
The environmental-science entries present component coefficients for engineering, limnology, horticulture, plant sciences, multidisciplinary categories, chemical technology, and public medical ethics, using principal component analysis with Varimax rotation.
- Environmental sciences: Engineering entries report coefficients 0.205, 0.398, 0.682 and 0.109, 0.562, 0.395 in separate rows.Additional rows report chemical technology as 0.362, 0.107, 0.373 and public medical ethics as 0.139, 0.721.
- Environmental sciences: Limnology has coefficients 0.541, 0.251, and 0.610 across the reported components.
- Environmental sciences: Horticulture has coefficients 0.119, 0.101, and 0.802 across the reported components.
- Environmental sciences: Plant sciences has coefficients 0.293, 0.223, and 0.714 across the reported components.
- Method: Principal Component Analysis was used for extraction, followed by Varimax rotation with Kaiser Normalization.The rotation converged in 11 iterations.