STRUCTURES OF SCIENTIFIC COOPERATION OF THREE RUSSIAN ORGANIZATIONS: SOCIOSEMANTIC NETWORK ANALYSIS

Patarakin E.D.

Moscow City Pedagogical University
129226, 2nd Agricultural Drive, 4
Moscow, Russia
E-mail: patarakined@mgpu.ru

Introduction

A comparative study of scientific organizations and schools through the lens of publication activity is of significant methodological interest for the sociology of science, science studies, and organizational psychology [2]. Bibliometric methods allow us to objectify the structure of scientific discourse, identifying both social (who interacts with whom) [1; 4] and semantic (what is written about) patterns. Combining these two analytical planes within the framework of sociosemantic analysis opens up opportunities for comparing not only the formal structures of scientific communities but also their conceptual spaces. Such an analysis is particularly relevant in relation to Russian psychology. The Moscow Psychological Society, founded in 1885, inherits the traditions of classical European psychology and simultaneously gave rise to its own scientific schools of global significance. In contemporary Russia, psychological knowledge is organized around several institutional centers with different traditions, objectives, and models of collaboration. This paper examines three leading centers of Russian psychology: the Russian Academy of Sciences (the Institute of Psychology of the Russian Academy of Sciences), Lomonosov Moscow State University (the Faculty of Psychology), and Moscow City Pedagogical University (MCPU). These three organizations represent different institutional types: an academic research institute with a focus on fundamental research, a classical university combining education and science, and a pedagogical university with a distinct applied and educational focus.

Methodologically, the study relies on the concept of teamwork [8; 11] as an integral characteristic of the degree of authors' involvement in stable cooperative networks, and on the sociosemantic approach [5; 10], which considers author networks in conjunction with the semantic content of publications. A key theoretical assumption is that social ties and semantic patterns coevolve. Authors who collaborate closely are highly likely to share conceptual space, and vice versa.

The main research question is formulated as follows: how do the structures of scientific collaboration and thematic profiles of psychologists from three leading Russian institutions differ, and what do these differences reveal about the organization of psychological knowledge in Russia?

The answer to this question allows us, on the one hand, to test the applicability of the classical concept of the «invisible college» [9] in the Russian context [3], and on the other hand, to propose a more detailed typology of scientific teamwork models.

Materials and methods

The source data was downloaded from the open bibliometric database Lens.org, an academic publication aggregator that indexes over 200 million scientific papers from Crossref, PubMed, Microsoft Academic, and patent registries. Unfortunately, we were unable to use the Russian Science Citation Index (RSCI) bibliometric database, as it does not offer open access via an open API.

Lens.org provides structured metadata in CSV format, including the list of authors, title, year of publication, edition, and citation data. The ability to filter authors by organizational affiliation allows for the reconstruction of local scientific communities without accessing internal institutional databases.

The study included three datasets representing the publication profiles of three Russian psychological schools.

The RAS dataset covers 270 articles by researchers at the Institute of Psychology of the Russian Academy of Sciences for the period 2007–2025. After splitting the author field, 671 rows were obtained, identifying 444 unique authors.

The Moscow State University dataset includes 979 publications for the period 2007–2024. After splitting the author field, 2,371 rows were obtained; 1,401 unique authors were identified.

The Moscow State Pedagogical Univ. dataset contains 59 publications for the period 2015–2024. After splitting the author field, 134 rows were obtained; 110 authors were identified.

Correct network analysis requires high-quality source data. A key challenge when working with bibliometric databases is the ambiguity of author identifiers: the same researcher may appear in different publications under different name variations—with full names or initials, in different transliterations, with typos, or missing diacritics. Without eliminating this ambiguity, the co-authorship network graph contains "phantom" nodes, and metrics such as density, centrality, and the size of the largest connected component become systematically distorted.

To standardize author identifiers, the OpenRefine environment, which implements the entity resolution approach, was used. The cleansing pipeline included five sequentially applied algorithms, alternating methods of varying degrees of aggressiveness: Fingerprint (Key Collision), N-Gram Fingerprint (Key Collision), Beider-Morse Phonetic Matching (Key Collision), Levenshtein Distance (Nearest Neighbor, threshold d ≤ 2), and PPM – Prediction by Partial Matching (Nearest Neighbor). The order was determined by the principle "from conservative procedures to more aggressive": methods with the lowest risk of false fusions were applied first, and only after these were exhausted were more aggressive algorithms employed. Each clustering operation required explicit confirmation by the researcher, precluding automated fusion without quality control.

As a result of using the data cleaning pipeline, the number of unique author records decreased: from 575 to 483 records (−16.0%) in the RAS dataset, from 121 to 112 (−7.4%) in the MGPU dataset, and from 1723 to 1413 (−18.0%) in the MSU dataset. The differences in deduplication rates are substantively significant: the relatively small scale of deduplication at MGPU (7.4%) reflects the greater homogeneity of the author base; the largest relative reduction in the MSU dataset (18.0%) indicates the wide variability of naming, typical of large aggregated datasets.

The co-authorship network was formally represented as a unimodal graph, where the set of nodes V are unique authors, and an edge exists if and only if authors i and j are co-authors of at least one publication; the edge weight is equal to the number of joint publications. This graph is a projection of the original bipartite "authors – publications" network. For each school, the following were calculated: global graph characteristics (number of nodes, edges, density, transitivity, number of components, size and proportion of the largest connected component, average path length in LCC), node characteristics (degree, weighted degree, betweenness centrality, closeness centrality, coreness), and team characteristics (average team size per publication, proportion of joint and team publications, distribution of team sizes). To compare degree distributions between schools, the two-sample Kolmogorov–Smirnov test and the nonparametric Kruskal–Wallis test with Dunn's post-hoc test were used. Author roles were classified along two dimensions—coreness and betweenness: "core-broker" (high coreness and high betweenness), "core" (high coreness, low betweenness), "broker" (low coreness, high betweenness), and "periphery" (both below the median). Thresholds were determined by the third quartile of the corresponding indicator within each school. Additionally, the teamness index was calculated—a normalized integral indicator aggregating z-standardized values of degree, stigmergy (the product of team size and the logarithm of the number of citations) [6], author activity duration, and average team size.

Publication titles were used for semantic analysis. Title texts were tokenized, converted to lowercase, and stop words were removed. A co-word network was constructed for each school: nodes represented unique terms, and edges represented their co-occurrence in a single title with a co-occurrence frequency of at least 2. The following parameters were calculated: the number of unique terms, the semantic network density, and the top 10 most frequent terms. For quantitative comparison of the schools' subject spaces, the Jaccard coefficient [7] was calculated for the top 50 terms for each pair of schools.

All calculations were performed in the R environment using the igraph, tidygraph, ggraph, tidytext, widyr, rstatix, and tidyverse packages. The final presentation of the co-authors' diagrams was done using the Graphviz environment.

Results

The average team size per publication is 2.27 authors at Moscow State Pedagogical Univ., 2.42 at Moscow State University, and 2.48 at the Russian Academy of Sciences. All three schools work predominantly in the small team format. The distribution of team sizes shows that at Moscow State University, the most frequent format is two-author publications (320 articles), followed by single-author papers (304) and publications with three (173) and four (99) authors. At the Russian Academy of Sciences, single and paired publications predominate (88 and 71, respectively), but large collaborations of up to 13 authors are also present. At Moscow State Pedagogical Univ., with a smaller total number of works, two-author articles (21) lead, followed by single-author papers (17), and only two publications have five authors. Thus, Moscow State Pedagogical Univ. and the Russian Academy of Sciences generate large author teams to a greater extent than Moscow State Pedagogical Univ. The structural analysis revealed fundamental differences in the network topology with a comparable co-authorship frequency (Table 1).

Table 1. Main network characteristics of three schools

Indicator	Moscow State Pedagogical Univ.	Moscow State University	RAS
Number of authors (nodes)	110	1401	444
Number of ribs	112	2479	796
Density	0,0187	0,00253	0,00809
Transitivity	0,874	0,591	0,808
Average degree	2,04	3,54	3,59
Median degree	2	3	3
Number of components	42	300	111
LCC size (authors)	7	422	29
LCC share, %	6,4	30,1	6,5
Average path length in LCC	~1,20	6,88	2,43
Average coreness	1,91	2,92	3,18
Average betweenness	0,0000371	0,000413	0,0000605

MGPU has the highest density (0.0187) and the highest transitivity coefficient (0.874), indicating closely connected local clusters. However, the network is extremely fragmented: 42 components with 110 authors, the largest with only 7 members (6.4% of the total). This is a classic pattern of "islands"—small, stable teams that are strongly connected internally but weakly interconnected.

RAS, with a similar LCC share (≈6.5%) and high transitivity of 0.808, also represents a set of locally dense clusters. However, the average node degree at RAS is higher (3.59 versus 2.04 at MSPU), and the average coreness is the highest of the three schools (3.18), indicating a deeper core within each cluster.

MSU is fundamentally different: with the lowest density (0.00253) and lower transitivity (0.591), it contains a gigantic component encompassing 422 authors, or 30.1% of the entire network. This demonstrates that MSU functions not as a set of isolated teams, but as an integrated field of co-authorship that encompasses a significant portion of the school.

The distribution of edge weights shows that the overwhelming majority of co-authorships in all three schools are one-off: 92.9% edges at MSPU, 88.7% at MSU, and 93.0% at RAS have a weight of 1 (one joint publication). However, MSU stands out with a slightly higher proportion of repeated ties: 11.3% edges have a weight ≥ 2 (versus 7.1% at MSPU and 7.0% at RAS), and 2.9% edges have a weight ≥ 3 (versus 1.8% and 2.1%). The strongest dyadic ties are also concentrated at MSU: individual pairs have published together up to 8 times, while the maximum edge weight at MSPU is 3, and at RAS – 4.

At Moscow State Pedagogical Univ., the most productive author has 7 publications; the leaders in degree centrality have modest values (the maximum degree is 6). Most authors with high degrees are embedded in small, dense components with very low betweenness, indicating a local coordinating role without structural brokerage. At the Russian Academy of Sciences, the most productive authors reach 11 publications, and the leaders in degree centrality have values of 12–13. However, many high-degree authors have betweenness values close to zero: they occupy central positions within specific clusters, without connecting different parts of the network. This is consistent with a segmented structure of several relatively autonomous cores. Moscow State University demonstrates a fundamentally different productivity pattern. Some authors have 28 publications, while the maximum number reaches 43. Crucially, these same authors also exhibit the highest betweenness centrality: for example, the most productive author combines a degree of 43 with a normalized betweenness of 0.0469, while the second most productive author combines a degree of 33 with a betweenness of 0.0345. This means that the most central actors at MSU are not simply productive co-authors, but they connect different parts of the network, acting as institutional brokers.

The aggregated team profile (Table 1) confirms that MSU researchers, on average, occupy smaller and more locally constrained positions, while MSU and RAS researchers are more deeply embedded in the network core. RAS has the highest average coreness score, supporting the hypothesis of strong local cores; however, the significantly smaller giant component and lower average betweenness score compared to MSU suggest that these cores remain more segmented.

The distribution of authors by team roles is particularly revealing. At Moscow State Pedagogical Univ., 67.3% authors were classified as brokers and 32.7% as "core brokers"; no authors were identified in "pure core" or "periphery" roles. This atypical pattern corresponds to a small, compact network, where most active authors act as intermediaries within or between small clusters, without forming a distinct hierarchy.

At Moscow State University and the Russian Academy of Sciences, the majority of authors are classified as peripheral (59.71 TP3T and 55.41 TP3T, respectively), while the proportion of "pure core" authors is 15.11 TP3T and 19.11 TP3T. This structure is typical of large scientific fields with a relatively limited core organizing collaboration around a broad peripheral base.

A thematic analysis of publication titles reveals substantive differences in the conceptual frameworks of the three schools. The Jaccard coefficient for the top 50 terms shows the greatest similarity between RAS and MSU (J = 0.321), followed by MGPU and MSU (J = 0.287), and the least between RAS and MSU (J = 0.181). The RAS's thematic profile reflects fundamental and theoretical areas of psychology: psychological, cognitive, and group-oriented concepts consistently dominate titles. MSU combines a broad thematic coverage with a pronounced presence of clinical psychology, psychological-pedagogical, and inclusive areas, consistent with the profile of a large university with diverse applied programs. MGPU demonstrates a more specific conceptual profile, gravitating toward educational psychology, competency-based approaches, and pedagogical technologies.

Semantic networks also differ in density: the most compact network of co-word relationships is found at MGPU, which is consistent with the smaller corpus size; the most extensive network is found at MSU, where the large volume of publications generates a wide network of conceptual co-orientations.

Discussion

The obtained results allow us to interpret the three schools considered as three different types of organizational and socio-semantic configuration of the scientific community.

MGU is most accurately characterized as a compact ecology of small, highly cohesive, and loosely interconnected teams. Collaboration is real and frequent, but almost exclusively localized. High transitivity with a minimal giant component size indicates that teams are closely interconnected within themselves (approaching cliques) but lack structural bridges to the outside world. The average path length in the LCC is approximately 1.20, meaning that virtually all authors in the largest cluster are direct co-authors. In terms of teamwork, local teamwork is high, but the school-wide "team field" is structurally fragmented. This corresponds to the image of an institution in which research activity unfolds within small, stable teams, each of which may have its own thematic focus, and where there are virtually no intermediary authors between teams. The co-authorship network at MGU is presented in Figure 1.

Fig. 1. Co-authorship network of Moscow State Pedagogical Univ.

The Russian Academy of Sciences (RAS) represents a different type: a medium-sized network with moderate density and significant fragmentation. The presence of 111 components, despite the relatively small size of the largest (29 authors), indicates a mosaic structure, where individual research groups are relatively autonomous. Unlike Moscow State University (MSU), no giant component serving as a "meta-community" has formed; instead, there is a set of dense local cores with a high level of internal coreness. High transitivity (0.808) indicates strong cluster cooperation, while low betweenness values for most central actors indicate that these clusters remain inward-looking. In terms of teamwork, the network exhibits strong teams and highly coherent local cores, but a less pronounced overall "highway" of the school. This configuration is typical of academic structures, where research laboratories or groups enjoy significant autonomy and form their own cognitive-social subsystems. The RAS co-authorship network is presented in Fig. 2.

Fig. 2. RAS co-authorship network.

MSU represents the third – and most institutionally integrated – type. A large, giant component (422 authors, 30.1% networks), several authors with a high degree/betweenness ratio, and a high proportion of repeated ties all point to the presence of a broad, integrated collaborative field, in which multiple teams are connected through a common structure. Author-brokers are not simply productive – they connect different parts of the network, ensuring the circulation of co-authorship and, likely, conceptual norms between subgroups. The average path length in the LCC (6.88), despite the significant number of components, indicates the long length of collaborative routes: researchers are connected to one another through chains of intermediaries, which corresponds to the image of a differentiated field with several «subcenters» of activity. A fragment of the giant co-authorship component at MSU is shown in Figure 3.

Fig. 3. MSU co-authorship network (giant component).

The concept of an "invisible college" describes informal elite networks of approximately 100 researchers who maintain intensive information exchange and establish norms for research orientation in their discipline. When applied to research data, this concept reveals both its explanatory power and limitations. For Moscow State University, the concept of an "invisible college" is particularly applicable: a gigantic component of 422 authors, several highly centralized actors-brokers, and stable, recurring dyadic connections create an infrastructure for the formation of shared discursive standards and linguistic norms, disseminated through the structure of co-authorship. The normative function of an "invisible college" can be realized at Moscow State University through the role of authors with high betweenness, connecting various subfields. At the Russian Academy of Sciences, the concept is only partially applicable: strong local cores can function as autonomous "invisible colleges" at the level of laboratories or research programs, but a unified "invisible college" does not form at the school level. The Russian academic context, where research groups enjoy a high degree of autonomy within institutional structures, gives rise to not one but several parallel such networks. At Moscow State Pedagogical Univ., the concept is least applicable: there is neither the characteristic size of an elite core nor multi-tiered integration. Instead, there are small, local teams that more closely resemble the model of a "research group" than an "invisible college" in the classical sense.

Thematic indicators complement the network picture. The high Jaccard coefficient between the Russian Academy of Sciences and Moscow State University (J = 0.321) indicates significant overlap of conceptual spaces: both schools work with similar sets of concepts, despite differences in their network architecture. Conversely, the lower value for the RAS–Moscow State Pedagogical Univ. pair suggests that Moscow State Pedagogical Univ. specializes in a more thematically specific set of areas. From a sociosemantic perspective, this means that different schools demonstrate different modes of coupling the social and semantic levels. At Moscow State University, a large core and author-brokers create the preconditions for the formation of common discursive standards. At the Russian Academy of Sciences, the multiplicity of local cores suggests greater thematic heterogeneity, where different groups maintain their own semantic subsystems. At Moscow State Pedagogical Univ., local micro-networks, loosely connected, likely generate an even more fragmented semantic landscape. A comparison of the teamwork indicators for the three scientific schools is shown in Figure 4.

Fig. 4. Comparative command of scientific schools.

An important methodological finding of the study is the demonstration that simply counting the share of co-authored publications is insufficient to characterize teamwork. Using this metric alone, the three schools appear quite similar (67–71% joint publications). However, network analysis reveals fundamental differences in the structural organization of collaboration. Teamwork is not limited to co-authorship frequency; it is also determined by whether collaboration is concentrated in isolated micro-teams, organized around segmented local cores, or integrated into a broader, more cohesive school network. The role of preliminary data cleaning is no less significant. Without author deduplication, density, component sizes, and degree distributions would be systematically distorted. In the RAS dataset before cleaning, up to 92 "phantom" nodes potentially existed, each of which generated false breaks in the LCC. After deduplication, the real co-authorship structure becomes visible—and it is this that theoretical interpretations are based on.

The analysis conducted has several limitations. First, all data is based on Lens.org indexing, which does not fully represent domestic Russian-language publications. Second, despite multi-stage deduplication, there remains the possibility of unresolved variant entries of the same author under different names. Third, article titles reflect only the top-level content of the publication and are subject to genre norms, limiting the depth of semantic analysis. Including abstracts, author keywords, and full texts would significantly enrich the semantic layer.

Literature

1. Dudko V.V., Patarakin E.D. Study of scientific schools of the university by means of bibliometric mapping // Territory of new opportunities. Bulletin of Vladivostok State University. 2023. Vol. 15. No. 1(65). P. 150–167. DOI: 10.24866/VVSU/2949-1258/2023-1/150-167.
2. Zhuravlev A.L., Kostrigin A.A. Quantitative analysis of journal publications of the Institute of Psychology of the Russian Academy of Sciences in the scientific field of moral psychology in 2002–2021. Part 1: bibliometric analysis // Institute of Psychology of the Russian Academy of Sciences. Social and Economic Psychology. 2024. Vol. 9. No. 1(33). P. 216–239. DOI: 10.38098/ipran.sep_2024_33_1_10.
3. Patarakin E.D., Burov V.V. «"Invisible College" of MES // Bulletin of Moscow City Pedagogical University. Series "Informatics and Informatization of Education". 2022. No. 2 (60). P. 38–52. DOI: 10.25688/2072-9014.2022.60.2.04.
4. Patarakin E.D., Kutuzov A.I., Dvoretskaya I.V. Multimodal learning analytics: bibliometric and ontological analysis // Education and Science. 2025. Vol. 27. No. 7. Pp. 33–71.
5. Basov N., Roth C. Socio-Semantic Networks // Handbook of Culture and Social Networks / ed. by N. Crossley, P. Widdop. Cheltenham: Edward Elgar Publishing, 2025. pp. 251–270. DOI: 10.4337/9781803928784.00027.
6. Bassanetti T., Cezera S., Delacroix M. et al. Cooperation and deception through stigmergic interactions in human groups // Proceedings of the National Academy of Sciences. 2023. Vol. 120.No. 42. Article e2307880120. DOI: 10.1073/pnas.2307880120.
7. Costa L. da F. Further Generalizations of the Jaccard Index [Electronic resource]. 2021. arXiv:2110.09619. URL: https://arxiv.org/abs/2110.09619 (date accessed: 02.05.2026).
8. DeChurch LA, Lungeanu A., Contractor NS. Think like a team: Shared mental models predict creativity and problem-solving in space analogs // Acta Astronautica. 2024. Vol. 214. pp. 701–711. DOI: 10.1016/j.actaastro.2023.10.022.
9. de Solla Price DJ, Beaver D. Collaboration in an invisible college // American Psychologist. 1966. Vol. 21.No. 11. pp. 1011–1018. DOI: 10.1037/h0024051.
10. Roth C., Cointet J.-P. Social and semantic coevolution in knowledge networks // Social Networks. 2010. Vol. 32.No. 1. pp. 16–29. DOI: 10.1016/j.socnet.2009.04.005.
11. Secchi D., Neumann M. Team Structure and Task Performance (Version 1.0.5) [Electronic resource] // CoMSES Computational Model Library. 2024. URL: https://www.comses.net/codebases/3a340ef1-1992-4a5a-8e09-5dac9f541720/releases/1.0.5/ (date accessed: 03.05.2026).

References

1. Dudko VV, Patarakin ED. A study of the university’s scientific schools using bibliometric mapping. Territory of new opportunities. Vestnik of Vladivostok State University, 2023, vol. 15, no. 1(65), pp. 150–167. DOI: 10.24866/VVSU/2949-1258/2023-1/150-167.
2. Patarakin ED, Kutuzov AI, Dvoretskaya IV. Multimodal academic analytics: bibliometric and ontological analysis. Education and Science, 2025, vol. 27, no. 7, pp. 33–72. DOI: 10.17853/1994-5639-2025-7-33-71.
3. Patarakin ED, Burov VV. «"Invisible College" MESH. Bulletin of Moscow City Pedagogical University. Series "Informatization and Informatization of Education", 2022, no. 2(60), pp. 38–52. DOI: 10.25688/2072-9014.2022.60.2.04.
4. Zhuravlev AL, Kostrigin AA. Quantitative analysis of journal publications of the Institute of Psychology of the Russian Academy of Sciences in the scientific direction of psychology of morality in 2002–2021. Part 1: bibliometric analysis. Institute of Psychology of the Russian Academy of Sciences. Social and Economic Psychology, 2024, vol. 9, no. 1(33), pp. 216–239. DOI: 10.38098/ipran.sep_2024_33_1_10.