ASGARI Yasaman
- Department of Mathematical Modeling and Machine Learning, University of Zurich, Zurich, Switzerland
- Clustering in networks, Collaboration in networks, Community structure in networks, Contact networks, Spreading, Temporal networks
Recommendations: 0
Review: 1
Website
https://www.yasasgari.com/
Areas of expertise
Graph theory, Machine learning on graphs, computational social science, science of science, temporal networks, community detection on temporal networks
Review: 1
14 May 2025
Measuring changes in polarisation using Singular Value Decomposition of network graphs
A new tool in the toolbox to measure polarisation in networks? A recommendation for “Measuring changes in polarisation using Singular Value Decomposition of network graphs”
Recommended by Mario Angst based on reviews by Yasaman Asgari and 2 anonymous reviewersSocietal polarisation, which S. Anastasi and G. D. Riva are concerned about in “Measuring changes
in polarisation using Singular Value Decomposition of network graphs” [1], has been a talking
point and sometimes diffuse object of concern for a while. In parallel, the issue has received sci-
entific attention from a variety of perspectives, ranging from political to complexity science and
from normative to empirical approaches [2]. From an empirically oriented, applied social science
perspective, a key area of discussion in the literature has been the measurement of polarization.
Polarisation is a concept, which can be critically approached, not least because the there can be
a danger of oversimplified normative orientations toward the value of social cohesion, which is
not necessarily some sort of ideal state for a functional democracy, and also because sometimes
the presence of conflict in a society may highlight unjust differences in material conditions [3].
Thus, measurement of polarization has to be put into context of how polarization is understood.
It is here, where S. Anastasi and G. D. Riva [1] add a crucial element to the discussion, by pro-
viding a suggestion for a measurement of polarization based on network data, and by shining a
light on how a geographical focus of the literature on polarization may have led to suboptimal
measurements. Seemingly, concern about polarization is an international phenomenon. However,
the actual literature, as S. Anastasi and G. D. Riva [1] convincingly argue, is heavily influenced by
research in the context of the United States of America, which has a legacy of seeing polarisation
solely through the prism of differences between two political parties.
S. Anastasi and G. D. Riva [1] are clear in their conceptual approach to polarisation. They focus
on a bi-modal (two-group) understanding of polarization, but argue for a diagnosis of increasing
polarization if two polarized groups diverge across different axes. In this way, their understanding
of polarization also connects to research that aims to distinguish differing or overlapping political
cleavages in societies [4] and is grounded by the researchers’ own understanding of the context
of New Zealand, where their research takes place.
In this context, the proposal both of utilizing network data and Singular Value Decomposition
(SVD) seems well justified and innovative, as is the suggestion to define the process of polarization
as “the loss of dimensionality of a graph observerd over time” [1, p.9]. Differently put, network
data can be unique in embedding cleavages in relational structure. And modeling social networks
through a graph embedding (Random Dot Product Graphs in the case of the paper), enables the
computation of both an optimal embedding dimension and SVD entropy (as a measure of network
complexity), which link well to the specific understanding of polarization.
S. Anastasi and G. D. Riva [1] demonstrate their approach with examples using both empirical and
simulated data. For their empirical case study, the rely on data from the social media platform pre-
viously knows as Twitter, showing an interesting finding of potentially increasing polarisation in
New Zealand climate debates between 2017-2020 and 2020-2023. Given that the social media plat-
form in question is not particularly useful for researchers as a data source on social interactions
anymore, this makes one think about potential for future studies using the method introduced in
the paper. S. Anastasi and G. D. Riva [1] provide the means to replicate their approach by means
of Julia code, which should give a good starting point for future studies (although reproducibility
would have been even higher through the provision of a way to reproduce the computational
environment, such as Docker or Nix). As the proposed approach is both elegant and widely gen-
eralizable to other networks, it would be fascinating to see it applied in different contexts.
Bibliography
[1] S. Anastasi and G. D. Riva, “The process of polarisation as a loss of dimensionality: measuring
changes in polarisation using Singular Value Decomposition of network graphs.” [Online].
Available: https://arxiv.org/abs/2403.18191
[2] S. A. Levin, H. V. Milner, and C. Perrings, “The dynamics of political polarization,” Proceedings
of the National Academy of Sciences of the United States of America, vol. 118, no. 50, Dec. 2021,
[3] D. Kreiss and S. C. McGregor, “A review and provocation: On polarization and platforms,”
New media & society, vol. 26, no. 1, pp. 556–579, Jan. 2024, https://doi.org/10.1177/14614448231161880.
[4] E. Borbáth, S. Hutter, and A. Leininger, “Cleavage politics, polarisation and participation
in Western Europe,” West European politics, vol. 46, no. 4, pp. 631–651, Jun. 2023, https://doi.org/10.1080/01402382.2022.2161786.