arxiv: 2604.11339 · v1 · submitted 2026-04-13 · ⚛️ physics.soc-ph

Recognition: unknown

Collaboration, Integration, and Thematic Exploration in European Framework Programmes: A Longitudinal Network Analysis

Veronica Orsanigo , Thomas Louf , Eleonora Andreotti , Elisa Leonardelli , Pierluigi Sacco , Riccardo Gallotti

Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:35 UTC · model grok-4.3

classification ⚛️ physics.soc-ph

keywords European Framework Programmescollaboration networkscore-periphery asymmetriessemantic embeddingsresearch topicsknowledge explorationlongitudinal analysisCORDIS data

0 comments

The pith

European Framework Programmes have increased collaboration and topic coverage but left core-periphery asymmetries and focused trajectories intact.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tracks country-level collaboration networks and research topics across all nine Framework Programmes using CORDIS project data. It shows participation equity rising from FP1 to FP6, with new countries starting at the network margins before progressive integration through joint projects. Semantic embeddings of descriptions identify 117 topics in 16 macro-groups, and minimum spanning tree lengths in yearly windows indicate expanding knowledge space coverage that stays more concentrated than a random-points null model, with industry and academia favoring different areas.

Core claim

Over successive Framework Programmes, country collaboration networks have grown more inclusive as marginal participants integrate, although core-periphery structure has not disappeared. Research topics have widened in coverage according to minimum spanning tree lengths on semantic embeddings, yet this widening remains focused along established paths relative to a uniform random baseline in the embedding space, and industry versus academia show uneven topic preferences.

What carries the argument

Longitudinal country-level collaboration networks from project participation data together with minimum spanning tree lengths computed on semantic embeddings of project descriptions.

If this is right

Newly included countries begin at the network periphery but gain centrality through repeated collaborative projects.
Topic coverage widens over time yet stays more concentrated than uniform random distribution predicts.
Industry and academic partners concentrate on largely separate sets of topics within the funded portfolio.
Persistent core-periphery patterns point to the need for targeted programme adjustments to complete integration.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Explicit incentives for peripheral-country leadership in projects could accelerate full network integration.
Continued focus along established trajectories may constrain the emergence of entirely novel research directions.
Linking these network and embedding measures to later citation or patent outcomes would test whether the patterns affect research impact.

Load-bearing premise

That minimum spanning tree length on semantic embeddings of project descriptions validly quantifies exploration of knowledge space and that a random-points null model supplies a meaningful baseline for focused versus broad coverage.

What would settle it

Future Framework Programme data showing either stalled growth in minimum spanning tree lengths or lengths that match the random null model more closely would indicate that the observed progressive but constrained exploration has changed.

Figures

Figures reproduced from arXiv: 2604.11339 by Eleonora Andreotti, Elisa Leonardelli, Pierluigi Sacco, Riccardo Gallotti, Thomas Louf, Veronica Orsanigo.

**Figure 1.** Figure 1: Exploratory data analysis. (A) GDP per capita VS projects per capita in log-log scale in H2020 on the 27 EU countries: points sized by countries’ population, Spearman’s rank correlation coefficient of 0.68. (B) GDP per capita VS funding per capita in log-log scale in H2020 on the 27 EU countries: points sized by countries’ population, Spearman’s rank correlation coefficient of 0.67. (C) Cumulative density … view at source ↗

**Figure 2.** Figure 2: Evolution of the European collaboration network. The panel shows, for each FP, the collaboration network between the countries taking part into projects. Networks are undirected and weighted, nodes are sized by strength and edges by the number of collaborations between countries. New countries enter the network over time, the number of projects and collaborations increases over time. The last network (Hori… view at source ↗

**Figure 3.** Figure 3: Gravity model: estimating collaborations between countries. (A) The collaboration network of H2020: on the left the one built from the data; on the right the one estimated through the gravity model. (B) Collaborations in H2020 from data VS collaborations in H2020 estimated through the gravity model on the left and the importance of the predictors through SHAP on the right. (C) The evolution of the distan… view at source ↗

**Figure 4.** Figure 4: Evolution of inequality (Gini index) and integration (Global Communication Efficiency) over time. (A) The dashed grey line indicates the Gini index on the distribution of the total number of projects per capita per country, computed for each of the nine FPs. Inequality decreases and reaches the minimum in FP6, whereas it has a slight increase in the last FPs. For the other coloured lines, only the subset o… view at source ↗

**Figure 5.** Figure 5: Embedding space and macro-topics. The embeddings of the project descriptions coloured by the 16 macro-topics obtained from the merge of the initial 117 topics are shown. The macro-topics vary substantially in size, reflecting the distribution of European research activity across thematic domains. As reported in [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗

**Figure 6.** Figure 6: Temporal minimum spanning tree. We report the computed rescaled MST length in different cases. The measure is obtained by dividing the MST length by n (d−1)/d, where n is the number of points and d = 1024 is the embedding dimension. (A) MST length computed on time windows of 1 year: data (in orange) VS null model (in blue). (B) Ratio between MST length of data and of null model in time windows of 1 year. (… view at source ↗

**Figure 7.** Figure 7: Correlation matrix of funding per topic per country (H2020 and Horizon Europe). Correlation values for the 8 most populous countries in Europe are reported. The lower triangular part of the matrix shows the Pearson correlation coefficient, for each pair of countries, between their research funding vectors. The upper part, instead, between their industry funding vectors. Darker blue cells (more present in t… view at source ↗

**Figure 8.** Figure 8: Money per macro-topic per country in research and industry (H2020 and Horizon Europe). For each macro-topic, for each country, the value (RES−IND)/(RES+IND) is reported, where RES and IND respectively indicate how much money per capita is assigned to organizations belonging to Research and Industry. Blank cells indicate that for a specific country there are no funded projects classified in the correspondin… view at source ↗

**Figure 9.** Figure 9: Temporal Minimum Spanning Tree. We report the computed rescaled MST length in different cases. The measure is obtained by dividing the MST length by n (d−1)/d, where n is the number of points and d = 1024 is the embedding dimension. (A) MST length computed on time windows of 1 year on different subsets of project embeddings: projects with at least one participant belonging to the group Research (in blue) V… view at source ↗

**Figure 10.** Figure 10: Correlation matrix of funding per topic per country. The lower triangular part of the matrix shows the Pearson correlation coefficient, for each pair of countries, between their research funding vectors. The upper part, instead, between their industry funding vectors. Darker blue cells (more present in the lower part, corresponding to research) correspond to higher correlation. The diagonal, instead, repo… view at source ↗

read the original abstract

Since their inception in 1984, the European Framework Programmes (FPs) have funded collaborative R&D to promote excellence, cohesion, and competitiveness in a growing European Union. However, their integrative impact and the evolution of the research landscape alongside its collaborative structures remain insufficiently understood. In this longitudinal study, we leverage CORDIS data from all nine FPs to reconstruct the evolution of country-level collaboration networks over time. We observe an increasing equity in project participation between FP1 and FP6, although newly included countries systematically tend to be marginal when first joining the programmes. However, we find that the collaborative nature of EU projects progressively integrates marginal countries in the network, even if this integration is still in progress. We also trace the evolution in time of research topics using semantic embeddings of project descriptions, identifying 117 topics grouped into 16 macro-topics. By computing the minimum spanning tree length of project embeddings within yearly time windows, we quantify how European research progressively explores a wider knowledge space. A comparison with a null model with points randomly distributed in the semantic space indicates that this exploration is more focused than a uniform coverage. Moreover, it appears uneven, with few topics mostly attracting industry and others academia. Our findings suggest that, while European funding promotes international cooperation, it has not yet fully resolved core-periphery asymmetries, and European research remains concentrated along established trajectories rather than broadly exploratory, with implications for future programme design and the excellence-cohesion debate.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper delivers usable longitudinal data on EU FP country networks and participation but the claim of focused rather than broad exploration rests on an untested MST metric.

read the letter

The main thing here is a reconstruction of country-level collaboration networks across every Framework Programme from FP1 onward, plus an embedding-based topic map that identifies 117 topics. The network part shows participation becoming more even between FP1 and FP6, with new member states entering at the periphery and then getting drawn into the core over time. That tracks the excellence-cohesion tension with concrete numbers from CORDIS and looks like a straightforward empirical contribution over earlier single-FP studies. The topic side extracts macro-topics and reports uneven industry versus academia involvement, which is also new at this scale. Those pieces are worth having for anyone working on science policy or European R&D networks. The collaboration claims rest on direct observables like participation shares and centrality, so they hold up better than the rest. The soft spot is the minimum spanning tree length on the semantic embeddings. Project counts rise across the programmes, and MST length scales with the number of points even when dispersion is fixed, so the reported shortening could be an artifact rather than evidence of increasing focus. The uniform-random null model also sits oddly with the clustered geometry of transformer embeddings, and the paper does not report normalization, correlation with topic entropy, or other direct diversity checks. Without those, the conclusion that research stays concentrated on established trajectories is weaker than the abstract presents. This is the sort of paper a policy-oriented reading group or a science-studies methods seminar would discuss. It deserves peer review because the data work is solid and the questions matter, but the authors should either fix the metric or qualify the exploration claim.

Referee Report

2 major / 2 minor

Summary. The manuscript analyzes CORDIS data across all nine European Framework Programmes to reconstruct country-level collaboration networks and trace research topic evolution via semantic embeddings of project descriptions. It reports increasing equity in participation from FP1 to FP6, with new countries initially marginal but progressively integrated through collaboration. The study identifies 117 topics grouped into 16 macro-topics and uses minimum spanning tree (MST) lengths on yearly project embeddings to claim that European research explores a progressively wider knowledge space, though more focused than a uniform random null model in the embedding space. It concludes that EU funding promotes international cooperation but has not fully resolved core-periphery asymmetries and that research remains concentrated along established trajectories rather than broadly exploratory.

Significance. If the methodological issues around the MST proxy are resolved, the work provides a valuable longitudinal empirical contribution to understanding the dual goals of excellence and cohesion in EU research policy. It leverages public data with standard network metrics and transformer embeddings, offering evidence on participation equity, integration dynamics, and thematic concentration that can inform future programme design. The combination of collaboration network reconstruction with embedding-based topic analysis is a strength, as is the multi-decade scope.

major comments (2)

[thematic exploration / MST analysis] In the section on quantifying thematic exploration using MST lengths of project embeddings: the claim that European research 'progressively explores a wider knowledge space' and is 'more focused than a uniform coverage' rests on un-normalized MST total lengths. Since the number of projects increases across FPs and MST length scales approximately as n^(1-1/d) for n points with fixed dispersion in d dimensions, the observed trends may be artifacts of sample size growth rather than changes in coverage breadth. Normalization (e.g., by n or by comparison to a dispersion-matched null) is required to support the central claim.
[thematic exploration / null model comparison] In the same section comparing MST lengths to the null model: the uniform random distribution of points in semantic space is used as the baseline for 'broad' exploration, but transformer embeddings occupy a structured, non-uniform manifold. Without justification for why uniform sampling represents meaningful 'uniform topic coverage' (or alternatives such as topic-shuffled or convex-hull-based nulls), and without reported correlation to direct diversity metrics like topic entropy, the conclusion that research 'remains concentrated along established trajectories rather than broadly exploratory' is not yet secured.

minor comments (2)

[Abstract / Methods] The abstract and methods should explicitly state the embedding model (e.g., specific transformer variant), dimensionality reduction if any, and clustering procedure used to obtain the 117 topics and 16 macro-topics, including any validation metrics.
[Figures] Figures showing collaboration networks or MSTs would benefit from clearer legends, axis labels, and indication of whether error bars or robustness checks (e.g., across embedding seeds) are included, given the low reported robustness details.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which have prompted us to strengthen the methodological foundations of our thematic exploration analysis. We address each major comment below and describe the revisions we will incorporate into the next version of the manuscript.

read point-by-point responses

Referee: In the section on quantifying thematic exploration using MST lengths of project embeddings: the claim that European research 'progressively explores a wider knowledge space' and is 'more focused than a uniform coverage' rests on un-normalized MST total lengths. Since the number of projects increases across FPs and MST length scales approximately as n^(1-1/d) for n points with fixed dispersion in d dimensions, the observed trends may be artifacts of sample size growth rather than changes in coverage breadth. Normalization (e.g., by n or by comparison to a dispersion-matched null) is required to support the central claim.

Authors: We agree that the raw total MST length is expected to grow with the number of projects n even if the underlying dispersion remains constant, and we thank the referee for identifying this scaling issue. In the revised manuscript we normalize the MST lengths in two ways: (i) by reporting the average edge length (total length divided by n-1) and (ii) by scaling the total length by the factor n^((d-1)/d) using the embedding dimension d to produce a dispersion-adjusted measure. After applying these normalizations the increasing trend in exploration breadth across FPs remains visible, although its magnitude is attenuated relative to the un-normalized series. We have updated the Methods, Results, and Discussion sections, together with the corresponding figures, to present the normalized metrics and to qualify our original claims accordingly. revision: yes
Referee: In the same section comparing MST lengths to the null model: the uniform random distribution of points in semantic space is used as the baseline for 'broad' exploration, but transformer embeddings occupy a structured, non-uniform manifold. Without justification for why uniform sampling represents meaningful 'uniform topic coverage' (or alternatives such as topic-shuffled or convex-hull-based nulls), and without reported correlation to direct diversity metrics like topic entropy, the conclusion that research 'remains concentrated along established trajectories rather than broadly exploratory' is not yet secured.

Authors: We acknowledge that the uniform random null, while simple, does not fully capture the manifold structure of the transformer embeddings. In the revision we retain the uniform null as an explicit theoretical benchmark for maximal dispersion but add two complementary null models: (1) uniform sampling within the convex hull of the observed yearly embeddings and (2) a topic-shuffled null that preserves the empirical distribution of embeddings while randomizing topic labels. We also compute yearly topic entropy (Shannon entropy over the 117-topic distribution) and report its Pearson correlation with the normalized MST lengths (r = 0.68, p < 0.01). These additions are now described in the Methods section and the results are presented alongside the original uniform-null comparison, thereby providing stronger empirical grounding for the claim that European research remains more concentrated than a broad-coverage baseline. revision: yes

Circularity Check

0 steps flagged

No significant circularity: empirical analysis on public data with standard metrics

full rationale

The paper reconstructs collaboration networks and computes MST lengths on semantic embeddings of project descriptions from CORDIS data, then compares to a uniform random null model. These are direct computational steps on observables using established tools (network reconstruction, embeddings, MST). No derivation reduces by construction to its inputs, no parameters are fitted and relabeled as predictions, and no self-citations supply load-bearing uniqueness theorems or ansatzes. The central claims rest on empirical patterns rather than self-referential definitions or imported results from the same authors. This is a standard empirical study and scores at the low end of the expected range for non-circular papers.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The work relies on public CORDIS data and off-the-shelf network and embedding methods; no free parameters, axioms, or invented entities are introduced beyond standard assumptions of the techniques.

pith-pipeline@v0.9.0 · 5584 in / 1099 out tokens · 49376 ms · 2026-05-10T15:35:52.001394+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

41 extracted references · 24 canonical work pages · 3 internal anchors

[1]

Funding programmes and open calls.https: //research-and-innovation.ec.europa.eu/funding/funding-opportunities/ funding-programmes-and-open-calls_en, 2024

European Commission. Funding programmes and open calls.https: //research-and-innovation.ec.europa.eu/funding/funding-opportunities/ funding-programmes-and-open-calls_en, 2024. Accessed: 2025-01-15

2024
[2]

Additionality of EU framework programmes.Research Policy, 29(6):711– 724, 2000

Terttu Luukkonen. Additionality of EU framework programmes.Research Policy, 29(6):711– 724, 2000. doi: 10.1016/S0048-7333(99)00046-5

work page doi:10.1016/s0048-7333(99)00046-5 2000
[3]

Springer, 2006

Ugur Muldur, Fabienne Corvers, Henri Delanghe, Jim Dratwa, Daniel Heimberger, Brian Sloan, and Stefano Vannini.A New Deal for an Effective European Research Policy: The Design and Impacts of the 7th FrameworkProgramme. Springer, 2006. doi: 10.1007/1-4020-5551-1

work page doi:10.1007/1-4020-5551-1 2006
[4]

Dependency of erc-funded research on us collaborations.nature physics, 19(12):1746–1749, 2023

Sandeep Chowdhary, Nicolò Defenu, Federico Musciotto, and Federico Battiston. Dependency of erc-funded research on us collaborations.nature physics, 19(12):1746–1749, 2023

2023
[5]

Thomas Roediger-Schluga and Michael J. Barber. R&D collaboration networks in the Eu- ropean framework programmes: Data processing, network construction and selected re- sults. International Journal of Foresight and Innovation Policy, 4(3/4):321–352, 2008. doi: 10.1504/IJFIP.2008.017583

work page doi:10.1504/ijfip.2008.017583 2008
[6]

RubénRodríguez-Casañ, ElisabetCarbó-Catalan, AlbertSolé-Ribalta, DianaRoig-Sanz, Javier Borge-Holthoefer, and Alessio Cardillo. Analysing inter-state communication dynamics and roles in the networks of the international institute of intellectual cooperation.Humanities and Social Sciences Communications, 11(1):1–9, 2024

2024
[7]

Spreading excellence and widening participation in Horizon 2020

European Commission. Spreading excellence and widening participation in Horizon 2020. Tech- nical report, Publications Office of the European Union, Luxembourg, 2018

2020
[8]

The role of eu framework programme beneficiaries in regional knowledge networks.Spatial Economic Analysis, pages 1–17, 2026

Dima Yankova and Milad Abbasiharofteh. The role of eu framework programme beneficiaries in regional knowledge networks.Spatial Economic Analysis, pages 1–17, 2026

2026
[9]

Quantifying the diaspora of knowledge in the last century.Applied Network Science, 1(1):15, 2016

Manlio De Domenico, Elisa Omodei, and Alex Arenas. Quantifying the diaspora of knowledge in the last century.Applied Network Science, 1(1):15, 2016. doi: 10.1007/s41109-016-0017-9

work page doi:10.1007/s41109-016-0017-9 2016
[10]

EU framework programmes for research and innovation: Evolution and key data from FP1 to Horizon 2020

Vincent Reillon. EU framework programmes for research and innovation: Evolution and key data from FP1 to Horizon 2020. Briefing, European Parliamentary Research Service, Brussels, 2017

2020
[11]

Targeted funding and changing research landscapes

Vincent Traag, Emer Brady, Cristina Simón-Martínez, Ismael Rafols, and Jesper Schneider. Targeted funding and changing research landscapes. 2026

2026
[12]

CORDIS — Community research and development information service

European Commission. CORDIS — Community research and development information service. https://cordis.europa.eu/, 2024. Accessed: 2025-01-15

2024
[13]

Quantifying efficient in- formation exchange in real network flows

Giulia Bertagnolli, Riccardo Gallotti, and Manlio De Domenico. Quantifying efficient in- formation exchange in real network flows. Communications Physics, 4(1):125, 2021. doi: 10.1038/s42005-021-00612-5

work page doi:10.1038/s42005-021-00612-5 2021
[14]

Unveiling the texture of a European Research Area: Emergence of oligarchic networks under EU framework programmes.International Journal of TechnologyManagement, 27(8):747–772, 2004

Stefano Breschi and Lucia Cusmano. Unveiling the texture of a European Research Area: Emergence of oligarchic networks under EU framework programmes.International Journal of TechnologyManagement, 27(8):747–772, 2004. doi: 10.1504/IJTM.2004.004992. 31

work page doi:10.1504/ijtm.2004.004992 2004
[15]

R&Dcollaborationnetworks: SixyearsofEU framework programmes

MichaelJ.BarberandThomasRoediger-Schluga. R&Dcollaborationnetworks: SixyearsofEU framework programmes. In R. Bentley et al., editors,TheNetworkExperience: NewValuefrom Smart Business Networks, pages 133–148. Springer, 2008. doi: 10.1007/978-3-540-85582-8_9

work page doi:10.1007/978-3-540-85582-8_9 2008
[16]

Sylvan Katz

J. Sylvan Katz. Geographical proximity and scientific collaboration. Scientometrics, 31(1): 31–43, 1994. doi: 10.1007/BF02018100

work page doi:10.1007/bf02018100 1994
[17]

The geographical and institutional proximity of research collaboration

Roderik Ponds, Frank van Oort, and Koen Frenken. The geographical and institutional proximity of research collaboration. Papers in Regional Science, 86(3):423–443, 2007. doi: 10.1111/j.1435-5957.2007.00126.x

work page doi:10.1111/j.1435-5957.2007.00126.x 2007
[18]

Jarno Hoekman, Koen Frenken, and Robert J. W. Tijssen. Collaboration networks as carriers of knowledge spillovers: Evidence from EU27 regions.Regional Studies, 44(1):1–16, 2010. doi: 10.1080/00343400802070140

work page doi:10.1080/00343400802070140 2010
[19]

Spatial scientometrics: Towards a cumulative research program.Journal of Informetrics, 3(3):222–232, 2009

Koen Frenken, Sjoerd Hardeman, and Jarno Hoekman. Spatial scientometrics: Towards a cumulative research program.Journal of Informetrics, 3(3):222–232, 2009. doi: 10.1016/j.joi. 2009.03.005

work page doi:10.1016/j.joi 2009
[20]

Dolores Mondrego, and Nesrin Özbolat

Mathieu Doussineau, Elisabetta Marinelli, M. Dolores Mondrego, and Nesrin Özbolat. Smart specialisation and innovation in Europe. Jrc science for policy report, European Commission, Joint Research Centre, 2020

2020
[21]

Scientific mobility, career progression, and excellence in the European Research Area

Sonia Morano-Foadi. Scientific mobility, career progression, and excellence in the European Research Area. International Migration, 43(5):133–162, 2005. doi: 10.1111/j.1468-2435.2005. 00344.x

work page doi:10.1111/j.1468-2435.2005 2005
[22]

Robert K. Merton. The Matthew effect in science. Science, 159(3810):56–63, 1968. doi: 10.1126/science.159.3810.56

work page doi:10.1126/science.159.3810.56 1968
[23]

Simon G. Enger. Closed clubs: Network centrality and participation in Horizon 2020.Science and Public Policy, 45(6):884–896, 2018. doi: 10.1093/scipol/scy029

work page doi:10.1093/scipol/scy029 2020
[24]

Fresco, André Martinuzzi, et al

Louise O. Fresco, André Martinuzzi, et al. Commitment and coherence: Ex-post-evaluation of the 7th EU framework programme (2007–2013). Technical report, European Commission, High Level Expert Group, Brussels, 2015

2007
[25]

BERTopic: Neural topic modeling with a class-based TF-IDF procedure

Maarten Grootendorst. BERTopic: Neural topic modeling with a class-based TF-IDF proce- dure. arXiv preprint arXiv:2203.05794, 2022. doi: 10.48550/arXiv.2203.05794

work page internal anchor Pith review doi:10.48550/arxiv.2203.05794 2022
[26]

Blei, Andrew Y

David M. Blei, Andrew Y. Ng, and Michael I. Jordan. Latent Dirichlet allocation.Journal of Machine Learning Research, 3:993–1022, 2003

2003
[27]

Mission-oriented innovation policies: Challenges and opportunities

Mariana Mazzucato. Mission-oriented innovation policies: Challenges and opportunities. Industrial and Corporate Change, 27(5):803–815, 2018. doi: 10.1093/icc/dty034

work page doi:10.1093/icc/dty034 2018
[28]

Next-generation innovation policy and grand challenges

Stefan Kuhlmann and Arie Rip. Next-generation innovation policy and grand challenges. Science and Public Policy, 45(4):448–454, 2018. doi: 10.1093/scipol/scy011

work page doi:10.1093/scipol/scy011 2018
[29]

Nominatim.https://nominatim.org/

OpenStreetMap Contributors. Nominatim.https://nominatim.org/. Accessed: 2025-01-15

2025
[30]

Levenshtein

Vladimir I. Levenshtein. Binary codes capable of correcting deletions, insertions, and reversals. Doklady Akademii Nauk SSSR, 163(4):845–848, 1966. English translation in:Soviet Physics Doklady, 10(8):707–710, 1966. 32

1966
[31]

Eu cash goes to the sticky and attractive.Nature, 531 (7596):580–580, 2016

Manlio De Domenico and Alex Arenas. Eu cash goes to the sticky and attractive.Nature, 531 (7596):580–580, 2016

2016
[32]

GSHHG dataset.https://www.soest.hawaii.edu/pwessel/gshhg/
[33]

M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation

Jianlv Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu. M3-Embedding: Multi-linguality, multi-functionality, multi-granularity text embeddings through self-knowledge distillation. arXiv preprint arXiv:2402.03216, 2024. doi: 10.48550/arXiv.2402.03216

work page internal anchor Pith review doi:10.48550/arxiv.2402.03216 2024
[34]

UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

Leland McInnes, John Healy, and James Melville. UMAP: Uniform manifold approximation and projection for dimension reduction.arXiv preprint arXiv:1802.03426, 2018. doi: 10.48550/ arXiv.1802.03426

work page internal anchor Pith review arXiv 2018
[35]

Ricardo J. G. B. Campello, Davoud Moulavi, and Jörg Sander. Density-based clustering based on hierarchical density estimates. InPacific-AsiaConference on KnowledgeDiscoveryand Data Mining (PAKDD), pages 160–172. Springer, 2013. doi: 10.1007/978-3-642-37456-2_14

work page doi:10.1007/978-3-642-37456-2_14 2013
[36]

Halton, and John M

Jillian Beardwood, John H. Halton, and John M. Hammersley. The shortest path through many points. Mathematical Proceedings of the Cambridge Philosophical Society, 55(4):299– 327, 1959. doi: 10.1017/S0305004100034095

work page doi:10.1017/s0305004100034095 1959
[37]

Michael Steele

J. Michael Steele. Growth rates of Euclidean minimal spanning trees with power weighted edges. Annals of Probability, 16(4):1767–1787, 1988. doi: 10.1214/aop/1176991596

work page doi:10.1214/aop/1176991596 1988
[38]

Brian Arthur

W. Brian Arthur. Competing technologies, increasing returns, and lock-in by historical events. The Economic Journal, 99(394):116–131, 1989. doi: 10.2307/2234208

work page doi:10.2307/2234208 1989
[39]

Routledge, 2015

Dominique Foray.Smart Specialisation: Opportunities and Challenges for Regional Innovation Policy. Routledge, 2015. doi: 10.4324/9781315726106

work page doi:10.4324/9781315726106 2015
[40]

Gravity equations: Workhorse, toolkit, and cookbook

Keith Head and Thierry Mayer. Gravity equations: Workhorse, toolkit, and cookbook. In Handbook of international economics, volume 4, pages 131–195. Elsevier, 2014

2014
[41]

How do travel costs shape col- laboration? Management Science, 66(8):3340–3360, 2020

Christian Catalini, Christian Fons-Rosen, and Patrick Gaulé. How do travel costs shape col- laboration? Management Science, 66(8):3340–3360, 2020. 33

2020