Angular separability of data clusters or network communities in geometrical space and its relevance to hyperbolic embedding

Alessandro Muscoloni; Carlo Vittorio Cannistraci

arxiv: 1907.00025 · v1 · pith:L3PJ7CNTnew · submitted 2019-06-28 · 💻 cs.LG · cs.SI· physics.soc-ph· stat.ML

Angular separability of data clusters or network communities in geometrical space and its relevance to hyperbolic embedding

Alessandro Muscoloni , Carlo Vittorio Cannistraci This is my paper

Pith reviewed 2026-05-25 13:28 UTC · model grok-4.3

classification 💻 cs.LG cs.SIphysics.soc-phstat.ML

keywords angular separation indexhyperbolic embeddingnetwork geometrycommunity detectionintrinsic dimensionalitygeometric embeddingcomplex networks

0 comments

The pith

The angular separation index quantifies community separation along angular coordinates and detects intrinsic network dimensionality.

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

The paper introduces the angular separation index (ASI) as a quantitative tool to assess how well node communities or data clusters separate along angular coordinates in a geometric embedding. A statistical test based on a uniform random null model determines if the observed separation is significant. This matters for big data analysis because embeddings in hyperbolic space are popular for representing hierarchical and similarity structures in networks, yet separation has mostly been checked visually. The authors use ASI to show that increasing temperature in 2D hyperbolic generative models not only lowers clustering but also makes the network behave as if it has more than two dimensions. Finally, ASI succeeds in recovering the intrinsic dimensionality of networks that were generated in a hidden geometric space.

Core claim

The angular separation index (ASI) is introduced to quantitatively evaluate the separation of network communities or data clusters over angular coordinates in any geometrical space, accompanied by an exact test statistic under a uniformly random null model; when applied to 2D hyperbolic embeddings it reveals that raising temperature induces a dimensionality jump beyond two and that ASI recovers the intrinsic dimensionality of networks grown in hidden geometry.

What carries the argument

Angular separation index (ASI), a statistic that evaluates the separation of labeled groups over angular coordinates with a significance test against uniform random placement.

If this is right

Increasing temperature in 2D hyperbolic generative models induces a jump to effective dimensions higher than two, in addition to reducing clustering.
ASI detects the intrinsic dimensionality of network structures that grow in a hidden geometrical space.
ASI supplies a statistically significant test for whether angular coordinates reflect community structure.
The index applies to angular separation assessment in any geometry, not only hyperbolic.

Where Pith is reading between the lines

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

ASI could serve as an objective validator for the quality of any angular embedding before using it for tasks like link prediction.
The dimensionality detection capability might guide selection of embedding dimension when the true geometry is unknown.
If ASI reliably links angular positions to communities, it would strengthen the interpretation of hyperbolic space as a natural representation for similarity data.

Load-bearing premise

The angular coordinates obtained from an embedding faithfully reflect the similarity or community structure present in the original data.

What would settle it

Computing ASI on networks with known communities and finding that the values are statistically indistinguishable from the uniform random null model would show that angular separation does not capture meaningful structure.

read the original abstract

Analysis of 'big data' characterized by high-dimensionality such as word vectors and complex networks requires often their representation in a geometrical space by embedding. Recent developments in machine learning and network geometry have pointed out the hyperbolic space as a useful framework for the representation of this data derived by real complex physical systems. In the hyperbolic space, the radial coordinate of the nodes characterizes their hierarchy, whereas the angular distance between them represents their similarity. Several studies have highlighted the relationship between the angular coordinates of the nodes embedded in the hyperbolic space and the community metadata available. However, such analyses have been often limited to a visual or qualitative assessment. Here, we introduce the angular separation index (ASI), to quantitatively evaluate the separation of node network communities or data clusters over the angular coordinates of a geometrical space. ASI is particularly useful in the hyperbolic space - where it is extensively tested along this study - but can be used in general for any assessment of angular separation regardless of the adopted geometry. ASI is proposed together with an exact test statistic based on a uniformly random null model to assess the statistical significance of the separation. We show that ASI allows to discover two significant phenomena in network geometry. The first is that the increase of temperature in 2D hyperbolic network generative models, not only reduces the network clustering but also induces a 'dimensionality jump' of the network to dimensions higher than two. The second is that ASI can be successfully applied to detect the intrinsic dimensionality of network structures that grow in a hidden geometrical space.

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

ASI is a usable new index for angular separation with a clean null model, but the dimensionality-jump and intrinsic-dimension claims rest on 2D embeddings whose distortion effects are not clearly separated from the signal.

read the letter

The paper introduces the Angular Separation Index (ASI) and pairs it with an exact test statistic under a uniform-random null model on the circle. That is the concrete addition: a quantitative replacement for the visual checks that have dominated work on angular coordinates in hyperbolic embeddings. The authors apply it to two observations in 2D hyperbolic generative models—temperature increase appears to push the effective dimension above two, and ASI recovers the hidden dimension of networks grown in an unseen geometry. Both are framed as new measurements enabled by the index. The index itself is straightforward to compute and the null model is external to the data, which keeps the circularity burden low. That part of the work is worth having on the shelf for anyone who needs to report angular separation with a p-value. The soft spot is the interpretation of the dimensionality results. The stress-test concern lands: when networks generated in D>2 are embedded back into 2D, the angular coordinates are necessarily projected, so any drop in ASI significance could be an artifact of that forced projection rather than evidence that the intrinsic dimension has changed. The abstract and the reported experiments do not describe controls that would disentangle embedding distortion from genuine higher-dimensional structure, such as direct comparison against known higher-D generators or against other dimension estimators. Without those, the two claimed phenomena remain suggestive but not yet isolated. The paper is aimed at the network-geometry and hyperbolic-embedding community. Readers who already work with angular coordinates will find a practical tool; readers looking for a settled method to detect intrinsic dimension will need the follow-up checks. It is coherent on its own terms and shows clear engagement with the literature, so it deserves a serious referee to examine the derivations and the experimental controls around the dimensionality claims.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces the Angular Separation Index (ASI) to quantitatively measure the angular separation of node communities or data clusters in a geometrical embedding (with emphasis on hyperbolic space), along with an exact statistical test against a uniformly random null model. It reports two main findings: increasing temperature in 2D hyperbolic generative models induces a dimensionality jump to dimensions higher than two, and ASI can detect the intrinsic dimensionality of networks that grow in a hidden geometrical space.

Significance. If the central claims survive scrutiny of embedding artifacts, ASI would supply a needed quantitative alternative to visual inspection of angular structure in embeddings, with direct relevance to network geometry and high-dimensional data analysis. The reported dimensionality phenomena could inform generative model design and embedding validation, provided they are shown to be independent of the 2D projection step.

major comments (2)

[Abstract and dimensionality-detection experiments] Abstract and the section describing ASI-based intrinsic-dimensionality detection: the central claim that ASI detects hidden dimension rests on computing ASI from angular coordinates obtained by embedding higher-D networks into fixed 2D hyperbolic space. Because any such embedding necessarily projects and mixes angular positions, a drop in ASI significance could be produced by the forced 2D projection itself rather than by the true intrinsic dimension; an explicit calibration that isolates embedding distortion from genuine dimensionality signal is required.
[Temperature-induced dimensionality jump] Section on temperature-induced dimensionality jump in 2D hyperbolic models: the reported jump is inferred from ASI values computed inside the same 2D embedding used to generate the networks. The ASI definition and its uniform-random null model are constructed for angular coordinates on a circle; without an auxiliary check that temperature-induced changes in embedding fidelity are not themselves driving the ASI decline, the dimensionality interpretation remains vulnerable to circularity.

minor comments (1)

The abstract supplies no derivation details, validation experiments, error analysis, or comparisons against alternative explanations; these elements should be made explicit in the main text so that readers can evaluate the load-bearing assumptions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed report. The two major comments correctly identify potential confounding effects from the 2D embedding step. We address each point below and will incorporate additional calibration experiments and checks in a revised manuscript to strengthen the claims.

read point-by-point responses

Referee: [Abstract and dimensionality-detection experiments] Abstract and the section describing ASI-based intrinsic-dimensionality detection: the central claim that ASI detects hidden dimension rests on computing ASI from angular coordinates obtained by embedding higher-D networks into fixed 2D hyperbolic space. Because any such embedding necessarily projects and mixes angular positions, a drop in ASI significance could be produced by the forced 2D projection itself rather than by the true intrinsic dimension; an explicit calibration that isolates embedding distortion from genuine dimensionality signal is required.

Authors: We agree that an explicit calibration isolating projection effects is required before the dimensionality-detection claim can be considered robust. In the revised version we will add a dedicated calibration subsection that (i) generates networks in known dimensions D=2,3,4 using the same hyperbolic model, (ii) embeds each into 2D hyperbolic space with the identical procedure used in the original experiments, and (iii) reports ASI together with standard embedding-quality metrics (stress, link-prediction AUC, angular reconstruction error). This will quantify how much of the observed ASI decline is attributable to projection distortion versus intrinsic dimensionality, allowing readers to assess the strength of the signal. revision: yes
Referee: [Temperature-induced dimensionality jump] Section on temperature-induced dimensionality jump in 2D hyperbolic models: the reported jump is inferred from ASI values computed inside the same 2D embedding used to generate the networks. The ASI definition and its uniform-random null model are constructed for angular coordinates on a circle; without an auxiliary check that temperature-induced changes in embedding fidelity are not themselves driving the ASI decline, the dimensionality interpretation remains vulnerable to circularity.

Authors: We acknowledge the circularity concern. Although the generative model is strictly 2D, temperature can affect both clustering and the ease of recovering the angular coordinates. In revision we will include, for each temperature, (i) the ASI value, (ii) the p-value from the uniform null model, and (iii) two embedding-fidelity diagnostics (average angular reconstruction error and the fraction of correctly recovered nearest neighbors in angle). If fidelity remains high while ASI drops, the dimensionality-jump interpretation is supported; if fidelity degrades, we will qualify the claim accordingly. These diagnostics will be reported in a new supplementary figure. revision: yes

Circularity Check

0 steps flagged

ASI definition and application show no load-bearing circularity; null model external and claims rest on independent generative tests

full rationale

The paper introduces ASI as a new quantitative index with an exact test based on a uniformly random null model stated as external to the data. No equations reduce ASI or its significance test to fitted parameters from the same embeddings, nor do central claims rely on self-citation chains for uniqueness or ansatz. The dimensionality-jump and intrinsic-dimension detection results are presented as empirical outcomes of applying ASI to 2D hyperbolic generative models and embeddings, without the derivation itself collapsing by construction to the inputs. Minor self-citation risk exists in the broader embedding literature but is not load-bearing here.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The central claim rests on the assumption that angular coordinates encode similarity and on the appropriateness of a uniform random null model; the index itself is a newly defined quantity.

axioms (2)

domain assumption Angular coordinates in the embedding represent similarity between nodes or clusters.
Stated in the abstract as the basis for using angular separation to evaluate communities.
domain assumption A uniformly random placement of labels provides a valid null model for testing angular separation significance.
Described in the abstract as the basis for the exact test statistic.

invented entities (1)

Angular Separation Index (ASI) no independent evidence
purpose: Quantitative score for angular separation of communities or clusters.
Newly introduced metric whose definition is not supplied in the abstract.

pith-pipeline@v0.9.0 · 5815 in / 1385 out tokens · 29043 ms · 2026-05-25T13:28:59.239562+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

the increase of temperature in 2D hyperbolic network generative models... induces a 'dimensionality jump' of the network to dimensions higher than two... ASI can be successfully applied to detect the intrinsic dimensionality of network structures that grow in a hidden geometrical space

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages · 2 internal anchors

[1]

Collective dynamics of ‘small -world’ networks,

D. J. Watts and S. H. Strogatz, “Collective dynamics of ‘small -world’ networks,” Nature, vol. 393, no. 6684, pp. 440–442, 1998

work page 1998
[2]

Power -Law Distributions in Empirical Data,

A. Clauset, C. Rohilla Shalizi, and M. E. J. Newman, “Power -Law Distributions in Empirical Data,” SIAM Rev., vol. 51, no. 4, pp. 661–703, 2009

work page 2009
[3]

An Information Flow Model for Conflict and Fission in Small Groups,

W. W. Zachary, “An Information Flow Model for Conflict and Fission in Small Groups,” J. Anthropol. Res., vol. 33, no. 4, pp. 452–473, 1977

work page 1977
[4]

Cross and A

R. Cross and A. Parker, The Hidden Power of Social Networks, no. October. 2004

work page 2004
[5]

Community structure in social and biological networks,

M. Girvan and M. E. J. Newman, “Community structure in social and biological networks,” PNAS, vol. 99, no. 12, pp. 7821–7826, 2002

work page 2002
[6]

The Political Blogosphere and the 2004 U.S. Election: Divided They Blog,

L. A. Adamic and N. Glance, “The Political Blogosphere and the 2004 U.S. Election: Divided They Blog,” LinkKDD 2005, pp. 36–43, 2005

work page 2004
[7]

Mach ine learning meets complex networks via coalescent embedding in the hyperbolic space,

A. Muscoloni, J. M. Thomas, S. Ciucci, G. Bianconi, and C. V. Cannistraci, “Mach ine learning meets complex networks via coalescent embedding in the hyperbolic space,” Nat. Commun., vol. 8, 2017

work page 2017
[8]

Latent Geometry Inspired Graph Dissimilarities Enhance Affinity Propagation Community Detection in Complex Networks

A. Muscoloni and C. V. Cannistraci, “Latent Geometry Inspired Graph Dissimilarities Enhance Affinity Propagation Community Detection in C omplex Networks,” arXiv:1804.04566 [cs.LG], 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018
[9]

Popularity versus similarity in growing networks,

F. Papadopoulos, M. Kitsak, M. A. Serrano, M. Boguñá, and D. Krioukov, “Popularity versus similarity in growing networks,” Nature, vol. 489, no. 7417, pp. 537–540, 2012

work page 2012
[10]

A nonuniform popularity-similarity optimization (nPSO) model to efficiently generate realistic complex networks with communities,

A. Muscoloni and C. V. Cannistraci, “A nonuniform popularity-similarity optimization (nPSO) model to efficiently generate realistic complex networks with communities,” New J. Phys., vol. 20, 2018

work page 2018
[11]

Leveraging the nonuniform PSO network model as a benchmark for performance evaluation in community detection and link prediction,

A. Muscoloni and C. V. Cannistraci, “Leveraging the nonuniform PSO network model as a benchmark for performance evaluation in community detection and link prediction,” New J. Phys., vol. 20, 2018

work page 2018
[12]

Local -community network automata modelling based on length -three-paths for predict ion of complex network structures in protein interactomes, food webs and more,

A. Muscoloni, I. Abdelhamid, and C. V. Cannistraci, “Local -community network automata modelling based on length -three-paths for predict ion of complex network structures in protein interactomes, food webs and more,” bioRxiv, 2018

work page 2018
[13]

Minimum curvilinear automata with similarity attachment for network embedding and link prediction in the hyperbolic space

A. Muscoloni and C. V. Cannistraci, “Minimum curvilinear automata with similarity attachment for network embedding and link prediction in the hyperbolic spa ce,” arXiv:1802.01183 [physics.soc-ph], 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018
[14]

Shortest Connection Networks And Some Generalizations,

R. C. Prim, “Shortest Connection Networks And Some Generalizations,” Bell Syst. Tech. J., vol. 36, no. 6, pp. 1389–1401, 1957

work page 1957
[15]

Network mapping by replaying hyperbolic growth,

F. Papadopoulos, C. Psomas, and D. Krioukov, “Network mapping by replaying hyperbolic growth,” IEEE/ACM Trans. Netw., vol. 23, no. 1, pp. 198–211, 2015

work page 2015
[16]

Sustaining the Internet with Hyperbolic Mapping,

M. Boguñá, F. Papadopoulos, and D. Krioukov, “Sustaining the Internet with Hyperbolic Mapping,” Nat. Commun., vol. 1, no. 6, pp. 1–8, 2010

work page 2010
[17]

Network Geometry Inference using Common Neighbors,

F. Papadopoulos, R. Aldecoa, and D. Krioukov, “Network Geometry Inference using Common Neighbors,” Phys. Rev. E, vol. 92, no. 2, p. 022807, 2015

work page 2015
[18]

Man ifold learning and maximum likelihood estimation for hyperbolic network embedding,

G. Alanis -Lobato, P. Mier, and M. A. Andrade -Navarro, “Man ifold learning and maximum likelihood estimation for hyperbolic network embedding,” Appl. Netw. Sci., vol. 1, no. 1, p. 10, 2016. N E m Cl γ C karate 34 78 2.3 0.59 2.1 2 opsahl 8 43 193 4.5 0.61 8.2 7 opsahl 9 44 348 7.9 0.68 5.9 7 opsahl 10 77 518 6.7 0.66 5.1 4 opsahl 11 77 1088 14.1 0.72 4.9...

work page 2016

[1] [1]

Collective dynamics of ‘small -world’ networks,

D. J. Watts and S. H. Strogatz, “Collective dynamics of ‘small -world’ networks,” Nature, vol. 393, no. 6684, pp. 440–442, 1998

work page 1998

[2] [2]

Power -Law Distributions in Empirical Data,

A. Clauset, C. Rohilla Shalizi, and M. E. J. Newman, “Power -Law Distributions in Empirical Data,” SIAM Rev., vol. 51, no. 4, pp. 661–703, 2009

work page 2009

[3] [3]

An Information Flow Model for Conflict and Fission in Small Groups,

W. W. Zachary, “An Information Flow Model for Conflict and Fission in Small Groups,” J. Anthropol. Res., vol. 33, no. 4, pp. 452–473, 1977

work page 1977

[4] [4]

Cross and A

R. Cross and A. Parker, The Hidden Power of Social Networks, no. October. 2004

work page 2004

[5] [5]

Community structure in social and biological networks,

M. Girvan and M. E. J. Newman, “Community structure in social and biological networks,” PNAS, vol. 99, no. 12, pp. 7821–7826, 2002

work page 2002

[6] [6]

The Political Blogosphere and the 2004 U.S. Election: Divided They Blog,

L. A. Adamic and N. Glance, “The Political Blogosphere and the 2004 U.S. Election: Divided They Blog,” LinkKDD 2005, pp. 36–43, 2005

work page 2004

[7] [7]

Mach ine learning meets complex networks via coalescent embedding in the hyperbolic space,

A. Muscoloni, J. M. Thomas, S. Ciucci, G. Bianconi, and C. V. Cannistraci, “Mach ine learning meets complex networks via coalescent embedding in the hyperbolic space,” Nat. Commun., vol. 8, 2017

work page 2017

[8] [8]

Latent Geometry Inspired Graph Dissimilarities Enhance Affinity Propagation Community Detection in Complex Networks

A. Muscoloni and C. V. Cannistraci, “Latent Geometry Inspired Graph Dissimilarities Enhance Affinity Propagation Community Detection in C omplex Networks,” arXiv:1804.04566 [cs.LG], 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018

[9] [9]

Popularity versus similarity in growing networks,

F. Papadopoulos, M. Kitsak, M. A. Serrano, M. Boguñá, and D. Krioukov, “Popularity versus similarity in growing networks,” Nature, vol. 489, no. 7417, pp. 537–540, 2012

work page 2012

[10] [10]

A nonuniform popularity-similarity optimization (nPSO) model to efficiently generate realistic complex networks with communities,

A. Muscoloni and C. V. Cannistraci, “A nonuniform popularity-similarity optimization (nPSO) model to efficiently generate realistic complex networks with communities,” New J. Phys., vol. 20, 2018

work page 2018

[11] [11]

Leveraging the nonuniform PSO network model as a benchmark for performance evaluation in community detection and link prediction,

A. Muscoloni and C. V. Cannistraci, “Leveraging the nonuniform PSO network model as a benchmark for performance evaluation in community detection and link prediction,” New J. Phys., vol. 20, 2018

work page 2018

[12] [12]

Local -community network automata modelling based on length -three-paths for predict ion of complex network structures in protein interactomes, food webs and more,

A. Muscoloni, I. Abdelhamid, and C. V. Cannistraci, “Local -community network automata modelling based on length -three-paths for predict ion of complex network structures in protein interactomes, food webs and more,” bioRxiv, 2018

work page 2018

[13] [13]

Minimum curvilinear automata with similarity attachment for network embedding and link prediction in the hyperbolic space

A. Muscoloni and C. V. Cannistraci, “Minimum curvilinear automata with similarity attachment for network embedding and link prediction in the hyperbolic spa ce,” arXiv:1802.01183 [physics.soc-ph], 2018

work page internal anchor Pith review Pith/arXiv arXiv 2018

[14] [14]

Shortest Connection Networks And Some Generalizations,

R. C. Prim, “Shortest Connection Networks And Some Generalizations,” Bell Syst. Tech. J., vol. 36, no. 6, pp. 1389–1401, 1957

work page 1957

[15] [15]

Network mapping by replaying hyperbolic growth,

F. Papadopoulos, C. Psomas, and D. Krioukov, “Network mapping by replaying hyperbolic growth,” IEEE/ACM Trans. Netw., vol. 23, no. 1, pp. 198–211, 2015

work page 2015

[16] [16]

Sustaining the Internet with Hyperbolic Mapping,

M. Boguñá, F. Papadopoulos, and D. Krioukov, “Sustaining the Internet with Hyperbolic Mapping,” Nat. Commun., vol. 1, no. 6, pp. 1–8, 2010

work page 2010

[17] [17]

Network Geometry Inference using Common Neighbors,

F. Papadopoulos, R. Aldecoa, and D. Krioukov, “Network Geometry Inference using Common Neighbors,” Phys. Rev. E, vol. 92, no. 2, p. 022807, 2015

work page 2015

[18] [18]

Man ifold learning and maximum likelihood estimation for hyperbolic network embedding,

G. Alanis -Lobato, P. Mier, and M. A. Andrade -Navarro, “Man ifold learning and maximum likelihood estimation for hyperbolic network embedding,” Appl. Netw. Sci., vol. 1, no. 1, p. 10, 2016. N E m Cl γ C karate 34 78 2.3 0.59 2.1 2 opsahl 8 43 193 4.5 0.61 8.2 7 opsahl 9 44 348 7.9 0.68 5.9 7 opsahl 10 77 518 6.7 0.66 5.1 4 opsahl 11 77 1088 14.1 0.72 4.9...

work page 2016