Pith. sign in

REVIEW 4 major objections 3 minor 61 references

CluStRE: Streaming Graph Clustering with Multi-Stage Refinement

T0 review · 4 major / 3 minor · reviewed 2026-08-08 · deepseek-v4-flash

Pith's one-line read A streaming graph clustering algorithm, CluStRE, claims to close the quality gap with in-memory methods by refining an initial node-stream clustering through evolutionary optimization on a quotient graph and re-streaming local search.

desk verdict A promising streaming clustering pipeline whose quotient-graph construction, as described, does not match the theorem used to justify it. read the letter →

arxiv 2502.06879 v1 pith:KI3SSZ55 submitted 2025-02-08 cs.LG cs.DB

classification cs.LGcs.DB
keywords graphclusteringcommunitydetectionstreamingalgorithmmodularityquotientmemeticre-streaminglocalsearch
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

CluStRE is a streaming graph clustering algorithm that tries to close the quality gap between streaming methods and in-memory methods like Louvain and VieClus. Its central proposal is to process the graph as a node stream, computing the locally best modularity gain for each node, then refine the result using two cheap global-information stages: evolutionary clustering on a dynamically built quotient graph, and re-streaming with local search. The paper reports that the strongest configuration reaches over 96% of Louvain's and VieClus's modularity on the tested graphs, while using less than a fifth of their memory, and that even the lightest configuration beats the streaming baseline Hollocou by about 90% modularity while running 2.6 times faster and using about 59% of its memory. If these numbers hold, the paper establishes that the streaming/in-memory quality trade-off is not as fundamental as often assumed.

What carries the argument

The load-bearing object is the quotient graph $G_Q$ built on the fly during streaming: each cluster becomes a weighted supernode, inter-cluster edges are aggregated into weighted edges, and intra-cluster edges become doubled self-loops. The paper proves (Theorem 1 in the appendix) that modularity of any clustering of $G_Q$ equals modularity of the corresponding clustering of the original graph, so refining the small quotient graph refines the original clustering. Two complementary refinement mechanisms exploit this: memetic (evolutionary) clustering, which recombines and mutates a population of quotient-graph clusterings, and re-streaming local search, which re-scans only the neighborhoods of nodes that changed cluster in the previous pass to accumulate modularity gain. The delta-modularity formula (Equation 3) drives every assignment decision.

What would settle it

Run the original Hollocou implementation, which randomizes edge order in memory, on the same benchmark set and compare modularity and memory against CluStRE; if the quality gap falls well below the reported 89.8% improvement or the memory advantage reverses, the central bridging claim is refuted.

Watch

Extended reading notes

Core claim

The paper proposes that the conventional quality ceiling of streaming graph clustering is an artifact of the one-pass constraint, not of streamed data itself. Its method, CluStRE, reads nodes one at a time and assigns each to the cluster that maximizes local modularity gain; then, in optional stages, it uses the current clustering to build a quotient graph on the fly and runs a memetic (evolutionary) clustering algorithm on that quotient graph, followed by re-streaming local search that revisits only the neighborhoods of changed nodes. The paper proves that modularity is exactly preserved under the quotient-graph contraction, so improvements found on the small graph are genuine improvements to the original. The reported outcome is that the strongest configuration reaches 96.8% of Louvain's and 96.5% of VieClus's modularity on average, while using 18.3% of Louvain's and 10.8% of VieClus's memory; the lightest configuration surpasses the Hollocou streaming baseline by 89.8% modularity, is 2.6 times faster, and uses 58.8% of its memory.

Load-bearing premise

The central numbers depend on the baseline streaming algorithm being fairly represented by the authors' modification that streams edges from disk in input order with the parameter set to 10,000; if that baseline or another streaming method were compared differently, the reported quality advantages could change.

Editorial extensions

If this is right

  • CluStRE-Strong can replace in-memory clustering on graphs too large for memory, losing only 3–4% modularity on average against Louvain and VieClus.
  • CluStRE-Light gives a strictly better streaming option than Hollocou: higher modularity, lower runtime, lower memory on the tested instances.
  • The quotient-graph modularity equivalence makes any modularity-based in-memory optimizer applicable to a streamed graph through CluStRE's on-the-fly contraction.
  • The active-node re-streaming scheme bounds I/O growth: only neighborhoods of reassigned nodes are re-read, so multiple refinement passes remain cheap.
  • Ground-truth community recovery, measured by NMI, also improves by about 17% over Hollocou, so the quality gain is not an artifact of the modularity objective alone.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural stress test is to replace the VieClus-based memetic optimizer on the quotient graph with a different modularity optimizer (e.g., Leiden) and re-measure the 96% figure; the quotient-graph invariance proof suggests the quality bridge is a property of the pipeline, not of one optimizer.
  • Because the paper's streaming baseline is a modified version of Hollocou that reads edges in input order, an editorially conservative check is to rerun the comparison with the original randomized-order implementation and with a second streaming algorithm; the 89.8% and 149.5% headline gaps are the most likely numbers to shift.
  • The memory measurements on uk-2007-05 (2.18 GB peak for the strongest mode) suggest that trillion-edge graphs become clusterable on commodity hardware; a direct follow-up is to scale the configuration curve to larger synthetic RGG and RHG instances and report the memory/quality crossover point.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 3 minor

Summary. The paper proposes CluStRE, a node-streaming graph clustering algorithm that assigns nodes by modularity gain and optionally refines the clustering by (i) memetic clustering on a quotient graph constructed during streaming and (ii) re-streaming local search. Four configurations are compared against a modified version of Hollocou's streaming algorithm and against Louvain and VieClus on large graphs, with claims of 89.8% quality improvement over Hollocou, 2.6x speedup, lower memory consumption, and over 96% of the modularity of in-memory methods. A quotient-graph modularity equivalence theorem is stated in the appendix.

Significance. If the empirical claims hold, CluStRE would be a useful practical contribution: it demonstrates that streaming clustering can approach in-memory modularity while using a fraction of the memory, and the four-mode design gives a transparent speed/quality trade-off. The paper is also careful in reporting resource usage and in describing algorithmic components. However, the central experimental claims rest on two unresolved issues: the on-the-fly quotient graph construction as written does not satisfy the theorem used to justify it, and the Hollocou baseline is modified and tuned in ways that may inflate the reported improvements. The paper would be publishable after a major revision that fixes the algorithmic description, re-runs the comparison against a fair baseline, and reports variance.

major comments (4)
  1. [§3.3, Algorithm 2; Appendix Theorem 1] Algorithm 2 constructs the quotient graph GQ while nodes are streamed, but it evaluates C[u] for each neighbor u at the moment the lower-ID endpoint v is processed. For every neighbor u that has not yet appeared in the stream, C[u] is still the initial singleton label; the algorithm never revisits or rewires these edges after u is later assigned to a non-singleton cluster. Consequently GQ contains supernodes that are not clusters of the final clustering, omits the self-loops that should represent internal edges of clusters formed by later-processed nodes, and in general is not the quotient graph of the final clustering C. The modularity equivalence in Appendix Theorem 1 is proved under the assumption that GQ is built from the final C, so the equivalence cannot be invoked for the graph actually built by Algorithm 2. The reported quality for CluStRE-Evo and CluStRE-Strong is therefore not attributable to the algorithm as written; the pseudocode needs an update rule that reroutes incident edges when a node joins a cluster (or a second construction pass after streaming).
  2. [§4, Baselines paragraph] The headline comparison to 'the state-of-the-art streaming clustering algorithm' is made against a modified Hollocou. The authors changed the edge order from the original randomized order to disk order, and selected vmax = 10,000 after testing multiple values because it gave the best quality. Since the paper itself states that Hollocou's quality is highly sensitive to edge-streaming order, this modification and parameter tuning can severely disadvantage the baseline; the reported 89.8% and 149.5% quality improvements and the 'best streaming method' conclusion are not supported without also reporting results for the original randomized order (or several orders) and for the parameter sensitivity of vmax.
  3. [§4.2, Table 1] The claim that CluStRE-Strong achieves 96.8% of Louvain and 96.5% of VieClus solution quality is averaged only over the instances on which the in-memory algorithms did not run out of memory (many rows in Table 1 are missing for these baselines). The abstract presents this as an unconditional statement. The paper should report the exact subset, per-instance quality ratios, and ideally also state what happens on the large instances where Louvain/VieClus fail; otherwise the 'over 96% of the quality of clustering approaches' claim is misleading.
  4. [§4, Methodology] All experiments appear to be single runs per instance, with no repeated trials, error bars, or variance reporting. This is problematic for the quantitative claims because Hollocou is order-sensitive, VieClus and CluStRE's memetic phase are stochastic, and the local-search time limit introduces run-to-run variation. The paper should report at least the range or standard deviation over several runs for the central quality, runtime, and memory comparisons.
minor comments (3)
  1. [§3.3, Algorithm 2] The key-handling in lines 9-12 of Algorithm 2 is ambiguous: if Q is meant to store undirected edges, the condition should be based on a canonical sorted pair; as written, when (Cj,Ci) exists but (Ci,Cj) does not, the code inserts a second entry and can double-count edge weights. Please clarify the canonical key representation.
  2. [§4, Experimental Setup and Reproducibility] The statement that code will be publicly available 'on acceptance' makes it impossible to verify the implementation against Algorithm 2; providing the artifact (or at least a detailed correction of the pseudocode) would strengthen reproducibility.
  3. [Figure 3 caption] The modularity performance profile in Figure 3(c) has a decreasing x-axis; the caption explains that higher modularity is better, but the axis direction should be labeled explicitly to avoid confusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the derivation is self-contained and the empirical claims are benchmarked externally. The skeptic's stale-assignment concern is a correctness issue, not circular reasoning.

full rationale

The paper's central claims are empirical benchmark results against external algorithms (Hollocou, Louvain, VieClus) and external graph datasets; no quantity that CluStRE outputs is defined in terms of the reported quality metric, and no reported quality number is a re-statement of a fitted parameter. The only formal derivation in the paper is Theorem 1 (Appendix A), which proves modularity invariance of the quotient graph by direct substitution from the definitions of modularity, intra-cluster edge weight, and volume (Eqs. 4-9). That proof assumes the quotient graph is built from the final clustering and derives the equivalence algebraically; it does not assume the conclusion. The streaming assignment uses the standard Louvain delta-modularity formula (Eq. 3), and the refinement steps are heuristics whose output is measured on held-out benchmark instances, not recovered from the input by construction. The paper does invoke VieClus, whose authorship overlaps with a co-author, and earlier buffered-streaming work by the same group, but those citations point to externally available algorithms and code used as a subcomponent and as related work; the central 96%-of-Louvain-quality claim is not justified by those citations alone. The only parameter tuned in the experimental section is vmax for the Hollocou baseline, and the only described tuning is a baseline-fairness choice that affects the comparison, not a CluStRE prediction derived from its own fit. The skeptic's observation that Algorithm 2 may use stale cluster assignments when constructing the quotient graph on the fly, if valid, would be an internal correctness/reproducibility flaw in the algorithm description rather than a circular step: it does not make the paper's output equivalent to its input by definition, nor does it rely on a self-citation to force the result. For these reasons, the appropriate circularity score is 0.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard modularity-based heuristics and on the choice of configuration parameters. No new physical or theoretical entities are postulated. The most consequential ledger entries are the hand-set time limits and thresholds, and the tuned Hollocou vmax that shapes the baseline comparison.

free parameters (5)
  • Hollocou vmax (baseline parameter) = 10,000
    Selected after testing multiple settings because it yielded the best clustering quality for Hollocou. This affects the magnitude of the claimed improvement over the streaming baseline (Section 4, Baselines).
  • Local search improvement threshold X = 0.05
    Hand-set parameter that stops re-streaming when a round's improvement falls below 5% of current modularity. Used in CluStRE-Light+ and CluStRE-Strong (Section 4, Experimental Setup).
  • Memetic refinement time limit = 15 seconds
    Hand-set cap on VieClus evolutionary rounds for CluStRE-Evo and CluStRE-Strong. Longer limits would increase quality at higher runtime cost (Section 4, Experimental Setup).
  • Local search time limit = 10 minutes
    Hand-set cap on the re-streaming local search phase. It influences both final quality and runtime (Section 4, Experimental Setup).
  • VieClus baseline time limit = 5 minutes
    Hand-set cap on the in-memory memetic baseline's evolutionary rounds. This affects the comparison of CluStRE quality relative to VieClus (Section 4, Experimental Setup).
assumptions (5)
  • domain assumption Modularity is an appropriate objective for graph clustering, and higher modularity indicates better clustering.
    The paper optimizes modularity and uses it as the main quality metric (Sections 1 and 2.1). This is a standard but contested assumption in community detection.
  • domain assumption Hollocou et al.'s algorithm is the current state of the art in streaming graph clustering.
    The paper benchmarks against this single streaming baseline and calls it state of the art (Section 4). If newer or stronger streaming methods exist, the headline comparison is incomplete.
  • domain assumption VieClus's memetic recombination and mutation operators reliably improve or preserve modularity on quotient graphs.
    CluStRE delegates refinement to VieClus, citing reference [7] for the property that recombined solutions are at least as good as their inputs (Section 3.3). The paper does not re-prove this.
  • domain assumption The graph is unweighted after preprocessing; all node and edge weights are set to 1.
    Section 4 states all graphs were converted to METIS format and unit weights assigned, so results may not transfer to weighted graphs.
  • standard math Modularity optimization is NP-complete, so heuristic algorithms are justified.
    Section 1 cites reference [12]. This is a standard complexity result that is not proven in the paper but is accepted background.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CluStRE: Streaming Graph Clustering with Multi-Stage Refinement." pith.science (2026). https://pith.science/paper/KI3SSZ55

@misc{pith2026250206879,
  author       = {Pith},
  title        = {Pith review of: CluStRE: Streaming Graph Clustering with Multi-Stage Refinement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KI3SSZ55}},
  note         = {Machine review of arXiv:2502.06879}
}
read the original abstract

We present CluStRE, a novel streaming graph clustering algorithm that balances computational efficiency with high-quality clustering using multi-stage refinement. Unlike traditional in-memory clustering approaches, CluStRE processes graphs in a streaming setting, significantly reducing memory overhead while leveraging re-streaming and evolutionary heuristics to improve solution quality. Our method dynamically constructs a quotient graph, enabling modularity-based optimization while efficiently handling large-scale graphs. We introduce multiple configurations of CluStRE to provide trade-offs between speed, memory consumption, and clustering quality. Experimental evaluations demonstrate that CluStRE improves solution quality by 89.8%, operates 2.6 times faster, and uses less than two-thirds of the memory required by the state-of-the-art streaming clustering algorithm on average. Moreover, our strongest mode enhances solution quality by up to 150% on average. With this, CluStRE achieves comparable solution quality to in-memory algorithms, i.e. over 96% of the quality of clustering approaches, including Louvain, effectively bridging the gap between streaming and traditional clustering methods.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages

  1. [1]

    of the 10th DIMACS Impl

    Proc. of the 10th DIMACS Impl. Challenge , Cont. Mathematics. AMS

  2. [2]

    Akoglu, H

    L. Akoglu, H. Tong, and D. Koutra. G raph B ased A nomaly D etection and D escription: A S urvey. Data Mining and Knowledge Discovery , 29(3):626--688, May 2015

  3. [3]

    Arnau, S

    V. Arnau, S. Mars, and I. Marı\' i n. I terative C luster A nalysis of P rotein I nteraction D ata. Bioinformatics , 21(3):364--378, September 2004

  4. [4]

    Assadi, V

    S. Assadi, V. Chatziafratis, V. Mirrokni, C. Wang, et al. H ierarchical C lustering in G raph S treams: S ingle-pass A lgorithms and S pace L ower B ounds. In Conf. on Learning Theory , pages 4643--4702. PMLR, 2022

  5. [5]

    D. A. Bader, H. Meyerhenke, P. Sanders, C. Schulz, A. Kappes, and D. Wagner. B enchmarking for G raph C lustering and P artitioning. In Encyclopedia of Social Network Analysis and Mining , pages 73--82. 2014

  6. [6]

    Bansal, A

    N. Bansal, A. Blum, and S. Chawla. C orrelation C lustering. Machine Learning , 56(1):89--113, Jul 2004

  7. [7]

    Biedermann, M

    S. Biedermann, M. Henzinger, C. Schulz, and B. Schuster. M emetic G raph C lustering. In 17th Intl. Symposium on Experimental Algorithms , volume 103 of Leibniz International Proceedings in Informatics (LIPIcs) , pages 3:1--3:15, 2018

  8. [8]

    V. D. Blondel, J. Guillaume, R. Lambiotte, and E. Lefebvre. F ast U nfolding of C ommunities in L arge N etworks. J. of Statistical Mechanics: Theory and Experiment , 2008(10):P10008, oct

Show all 61 references
  1. [9]

    Boldi, A

    P. Boldi, A. Marino, M. Santini, and S. Vigna. B U bi N G : M assive C rawling for the M asses. ACM Trans. Web , 12(2), 2018

  2. [10]

    Boldi, M

    P. Boldi, M. Rosa, M. Santini, and S. Vigna. L ayered L abel P ropagation: A M ultiresolution C oordinate-free O rdering F or C ompressing S ocial N etworks. In Proceedings of the 20th Intl. Conf. on World Wide Web , pages 587--596, 2011

  3. [11]

    Boldi and S

    P. Boldi and S. Vigna. T he W eb G raph F ramework I : C ompression T echniques. In Proc. of the Thirteenth Intl. World Wide Web Conf. , pages 595--601, 2004

  4. [12]

    Brandes, D

    U. Brandes, D. Delling, M. Gaertler, R. Görke, M. Hoefer, Z. Nikoloski, and D. Wagner. O n M odularity C lustering. IEEE Trans. Knowl. Data Eng. , 20(2):172--188, 2008

  5. [13]

    Chhabra, M

    A. Chhabra, M. F. Faraj, C. Schulz, and D. Seemaier. B uffered S treaming E dge P artitioning. In 22nd Intl. Symposium on Experimental Algorithms , volume 301 of LIPIcs , pages 5:1--5:21, 2024

  6. [14]

    Chhabra, F

    A. Chhabra, F. Kurpicz, C. Schulz, D. Schweisgut, and D. Seemaier. P artitioning T rillion E dge G raphs on E dge D evices, 2024

  7. [15]

    Dean and S

    J. Dean and S. Ghemawat. M ap R educe: S implified D ata P rocessing on L arge C lusters. Commun. ACM , 51(1):107–113, January 2008

  8. [16]

    Demme and S

    J. Demme and S. Sethumadhavan. A pproximate G raph C lustering for P rogram C haracterization. ACM Trans. Archit. Code Optim. , 8(4), January

  9. [17]

    Dhulipala, J

    L. Dhulipala, J. Łącki, J. Lee, and V. Mirrokni. T era H A C : H ierarchical A gglomerative C lustering of T rillion-edge G raphs. Proc. ACM Manag. Data , 1(3), November 2023

  10. [18]

    Diwan, Sanjeeva Rane, S

    Ajit A. Diwan, Sanjeeva Rane, S. Seshadri, and S. Sudarshan. C lustering T echniques for M inimizing E xternal P ath L ength. In T. M. Vijayaraman, Alejandro P. Buchmann, C. Mohan, and Nandlal L. Sarda, editors, VLDB'96, Proceedings of 22th International Conference on Very Lar...

  11. [19]

    E. D. Dolan and J. J. Moré. B enchmarking O ptimization S oftware W ith P erformance P rofiles. Math. Program. , 91(2):201--213, 2002

  12. [20]

    B uffered S treaming G raph P artitioning

    Marcelo Fonseca Faraj and Christian Schulz. B uffered S treaming G raph P artitioning. ACM J. Exp. Algorithmics , 27:1.10:1--1.10:26, 2022. https://doi.org/10.1145/3546911 doi:10.1145/3546911

  13. [21]

    Fortunato

    S. Fortunato. C ommunity D etection in G raphs. Physics Reports , 486(3):75--174, 2010

  14. [22]

    Funke, S

    D. Funke, S. Lamm, U. Meyer, M. Penschuck, P. Sanders, C. Schulz, D. Strash, and M. von Looz. C ommunication-free M assively D istributed G raph G eneration. J. Parallel Distributed Comput. , 131:200--217, 2019

  15. [23]

    Gaertler

    M. Gaertler. C lustering , pages 178--215. Springer Berlin Heidelberg, 2005

  16. [24]

    Girvan and M

    M. Girvan and M. E. J. Newman. C ommunity S tructure in S ocial and B iological N etworks. Proceedings of the National Academy of Sciences , 99(12):7821–7826, June 2002

  17. [25]

    M. R. Hajidehi, S. Sridhar, and M. I. Seltzer. C U T T A N A : S calable G raph P artitioning for F aster D istributed G raph D atabases and A nalytics. Proceedings of the VLDB Endowment , 18(1):14--27, 2024

  18. [26]

    Hamann, B

    M. Hamann, B. Strasser, D. Wagner, and T. Zeitz. D istributed G raph C lustering U sing M odularity and M ap E quation. In Euro-Par 2018: Parallel Processing , page 688–702, 2018

  19. [27]

    Hollocou, J

    A. Hollocou, J. Maudet, T. Bonald, and M. Lelarge. A S treaming A lgorithm for G raph C lustering. CoRR , abs/1712.04337, 2017

  20. [28]

    Jiang, S

    R. Jiang, S. Han, Y. Yu, and W. Ding. A n A ccess C ontrol M odel for M edical B ig D ata B ased O n C lustering and R isk. Information Sciences , 621:691--707, 2023

  21. [29]

    Kannan, S

    R. Kannan, S. Vempala, and A. Vetta. O n C lusterings: G ood, B ad and S pectral. J. ACM , 51(3):497–515, May 2004

  22. [30]

    Karypis and V

    G. Karypis and V. Kumar. A F ast and H igh Q uality M ultilevel S cheme for P artitioning I rregular G raphs. SIAM J. Sci. Comput. , 20(1):359--392, 1998

  23. [31]

    Lancichinetti, S

    A. Lancichinetti, S. Fortunato, and J. Kertész. D etecting the O verlapping and H ierarchical C ommunity S tructure in C omplex N etworks. New J. of Physics , 11(3):033015, March 2009

  24. [32]

    Lee and D

    C. Lee and D. J. Wilkinson. A R eview of S tochastic B lock M odels and E xtensions F or G raph C lustering. Applied Network Science , 4(1):122, Dec 2019

  25. [33]

    Leskovec

    J. Leskovec. S tanford N etwork A nalysis P ackage ( S nap), 2013

  26. [34]

    J. Li, S. Lai, Z. Shuai, Y. Tan, Y. Jia, M. Yu, Z. Song, X. Peng, Z. Xu, Y. Ni, H. Qiu, J. Yang, Y. Liu, and Y. Lu. A C omprehensive R eview of C ommunity D etection I n G raphs. Neurocomputing , 600:128169, 2024

  27. [35]

    F. Liu, S. Xue, J. Wu, C. Zhou, W. Hu, C. Paris, S. Nepal, J. Yang, and P. S. Yu. D eep L earning for C ommunity D etection: P rogress, C hallenges and O pportunities. In Proceedings of the Twenty-Ninth Intl. Joint Conf. on Artificial Intelligence , pages 4981--4987, 2020

  28. [36]

    Y. Liu, J. Xia, S. Zhou, X. Yang, K. Liang, C. Fan, Y. Zhuang, S. Z. Li, X. Liu, and K. He. A S urvey of D eep G raph C lustering: T axonomy, C hallenge, A pplication, and O pen R esource, 2023

  29. [37]

    Malewicz, M

    G. Malewicz, M. H. Austern, A. J. Bik, J. C. Dehnert, I. Horn, N. Leiser, and G. Czajkowski. P regel: A S ystem for L arge-scale G raph P rocessing. In Proceedings of the 2010 ACM SIGMOD Intl. Conf. on Management of Data , page 135–146, 2010

  30. [38]

    McFarling

    S. McFarling. P rogram O ptimization for I nstruction C aches. SIGARCH Comput. Archit. News , 17(2):183–191, April 1989

  31. [39]

    Meyerhenke, P

    H. Meyerhenke, P. Sanders, and C. Schulz. P artitioning C omplex N etworks via S ize-constrained C lustering. In Experimental Algorithms , pages 351--363, 2014

  32. [40]

    B. L. Miller and D. E. Goldberg. G enetic A lgorithms, S election S chemes, and the V arying E ffects of N oise. Evolutionary Computation , 4(2):113--131, 06 1996

  33. [41]

    M. E. J. Newman. P roperties of H ighly C lustered N etworks. Physical Review E , 68(2), August 2003

  34. [42]

    M. E. J. Newman. F ast A lgorithm for D etecting C ommunity S tructure I n N etworks. Physical Review E , 69(6), June 2004

  35. [43]

    M. E. J. Newman. M odularity and C ommunity S tructure in N etworks. Proceedings of the National Academy of Sciences , 103(23):8577–8582, June 2006

  36. [44]

    M. E. J. Newman and M. Girvan. F inding and E valuating C ommunity S tructure I n N etworks. Physical Review E , 69(2), February 2004

  37. [45]

    Nishimura and J

    J. Nishimura and J. Ugander. R estreaming G raph P artitioning: S imple V ersatile A lgorithms for A dvanced B alancing. In The 19th ACM SIGKDD Intl. Conf. on Knowledge Discovery and Data Mining , pages 1106--1114, 2013

  38. [46]

    J. B. Pereira-Leal, A. J. Enright, and C. A. Ouzounis. D etection of F unctional M odules F rom P rotein I nteraction N etworks. Proteins: Structure, Function, and Bioinformatics , 54(1):49--57, 2004

  39. [47]

    R. A. Rossi and N. K. Ahmed. T he N etwork D ata R epository with I nteractive G raph A nalytics and V isualization. In Proceedings of the Twenty-Ninth AAAI Conf. on Artificial Intelligence , pages 4292--4293, 2015

  40. [48]

    Rosvall, D

    M. Rosvall, D. Axelsson, and C. T. Bergstrom. T he M ap E quation. The European Physical J. Special Topics , 178(1):13--23, Nov 2009

  41. [49]

    S. Sakr, F. M. Orakzai, I. Abdelaziz, and Z. Khayyat. L arge-Scale G raph P rocessing U sing A pache G iraph . Springer Publishing Company, 2016

  42. [50]

    S. E. Schaeffer. G raph C lustering. Computer Science Review , 1(1):27--64, 2007

  43. [51]

    Shchur, M

    O. Shchur, M. Mumme, A. Bojchevski, and S. Günnemann. P itfalls of G raph N eural N etwork E valuation, 2018

  44. [52]

    X. Su, S. Xue, F. Liu, J. Wu, J. Yang, C. Zhou, W. Hu, C. Paris, S. Nepal, D. Jin, Q. Z. Sheng, and P. S. Yu. A C omprehensive S urvey on C ommunity D etection W ith D eep L earning. IEEE Transactions on Neural Networks and Learning Systems , 35(4):4682--4702

  45. [53]

    V. A. Traag, L. Waltman, and N. J. van Eck. F rom L ouvain to L eiden: guaranteeing well-connected communities. Scientific Reports , 9(1):5233, Mar 2019

  46. [54]

    C. E. Tsourakakis, C. Gkantsidis, B. Radunovic, and M. Vojnovic. F E N N E L : S treaming G raph P artitioning for M assive S cale G raphs. In Seventh ACM Intl. Conf. on Web Search and Data Mining , pages 333--342, 2014

  47. [55]

    Van Dongen

    S. Van Dongen. G raph C lustering V ia a D iscrete U ncoupling P rocess. SIAM J. on Matrix Analysis and Applications , 30(1):121--141, 2008

  48. [56]

    von Luxburg

    U. von Luxburg. A T utorial on S pectral C lustering. Statistics and Computing , 17(4):395--416, Dec 2007

  49. [57]

    Z. Wan, Y. Zhang, and H. He. V ariational A utoencoder B ased S ynthetic D ata G eneration for I mbalanced L earning. In 2017 IEEE Symposium Series on Computational Intelligence (SSCI) , pages 1--7, 2017

  50. [58]

    S. Wang, J. Yang, J. Yao, Y. Bai, and W. Zhu. A n O verview of A dvanced D eep G raph N ode C lustering. IEEE Transactions on Computational Social Systems , 11(1):1302--1314, 2024

  51. [59]

    Watteau, A

    T. Watteau, A. Bonnefoy, S. Illouz-Laurent, J. Jusseau, and S. Iovleff. A dvanced G raph C lustering M ethods: A C omprehensive and I n-depth A nalysis, 2024

  52. [60]

    Y. Xu, V. Olman, and D. Xu. C lustering G ene E xpression D ata U sing A G raph-theoretic A pproach: A n A pplication of M inimum S panning T rees. Bioinformatics , 18(4):536--545, 04 2002

  53. [61]

    Z. Yang, W. W. Cohen, and R. Salakhutdinov. R evisiting S emi-supervised L earning W ith G raph E mbeddings. In Proceedings of the 33rd International Conference on International Conference on Machine Learning , page 40–48, 2016

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.