pith. sign in

arxiv: 2509.21838 · v1 · submitted 2025-09-26 · 💻 cs.SI

Attributed Hypergraph Generation with Realistic Interplay Between Structure and Attributes

Jaewan Chun , Seokbum Yoon , Minyoung Choe , Geon Lee , Kijung Shin This is my paper

Pith reviewed 2026-05-18 13:25 UTC · model grok-4.3

classification 💻 cs.SI
keywords attributed hypergraphshypergraph generationcore-fringe hierarchystructure-attribute interplaystochastic modelsgenerative modelssocial hypergraphs
0
0 comments X

The pith

NoAH generates attributed hypergraphs by modeling hyperedge formation as sequential node attachments in a core-fringe hierarchy driven by node attributes.

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

The paper introduces NoAH, a generative model that treats hyperedge creation in attributed hypergraphs as a stochastic process of attaching nodes according to a core-fringe hierarchy, with probabilities set by each node's attributes. This setup aims to capture how real group interactions reflect both structural patterns and attribute influences, which previous models often overlooked. When combined with the NoAHFit procedure to learn parameters from data, the model is shown through tests on nine datasets to better match observed structure-attribute relationships than eight other generative approaches across six evaluation metrics. A reader might care because accurate synthetic hypergraphs can support simulation of social, biological, or collaboration networks where attributes like roles or features shape group formations. If the approach holds, it provides a way to create realistic test data that preserves these intertwined properties.

Core claim

NoAH is a stochastic hypergraph generative model for attributed hypergraphs. It utilizes the core-fringe node hierarchy to model hyperedge formation as a series of node attachments and determines attachment probabilities based on node attributes. NoAHFit is introduced as a parameter learning procedure that allows NoAH to replicate a given real-world hypergraph. Experiments demonstrate that NoAH with NoAHFit more accurately reproduces the structure-attribute interplay observed in real-world hypergraphs than eight baseline models, according to six metrics.

What carries the argument

Core-fringe node hierarchy for modeling hyperedge formation as stochastic node attachments with attribute-determined probabilities

Load-bearing premise

Hyperedge formation in attributed hypergraphs can be accurately captured by modeling it as a stochastic series of node attachments within a core-fringe hierarchy whose probabilities are determined solely by node attributes.

What would settle it

Observing that on additional real-world datasets, the fitted NoAH model fails to match the real hypergraphs on the structure-attribute metrics even when parameters are optimized.

Figures

Figures reproduced from arXiv: 2509.21838 by Geon Lee, Jaewan Chun, Kijung Shin, Minyoung Choe, Seokbum Yoon.

Figure 1
Figure 1. Figure 1: Structure–attribute interplay with respect to the first node attribute in the Amazon Music dataset. Our proposed generative model, NOAH, fitted by its parameter learning algorithm NOAHFIT, effectively captures the interplay between structure and node attributes, outperforming the baseline model (HyperCL). Refer to Section III-B for the definitions of the measures and to Section VI-C for the interpretation … view at source ↗
Figure 2
Figure 2. Figure 2: Hyperedge generation process of NOAH consists of two steps: (1) Core group construction: sample a seed core node vs according to pseed and attach additional core nodes to vs to form a core group Ce˜, and (2) Fringe attachment: attach fringe nodes to core group Ce˜ based on mixed attribute vector xCe˜ obtained by attribute-wise sampling from XCe˜ . The attached core and fringe nodes, together with the seed … view at source ↗
Figure 3
Figure 3. Figure 3: θ (1) C and θ (1) F estimated by NOAHFIT on the Amazon Music dataset. NOAHFIT captures homophily by assigning higher affinities to same attribute value pairs (0 ↔ 0 and 1 ↔ 1) than to different attribute value pairs (0 ↔ 1). of NOAH in capturing the structure–attribute interplay of real￾world hypergraphs. However, there are cases where NOAH underperforms. First, its relatively low performance in the online… view at source ↗
Figure 4
Figure 4. Figure 4: Structural patterns of real-world hypergraphs ( [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: NOAHFIT scales nearly linearly with the number of hyperedges and remains almost constant with respect to the number of attributes. NOAH scales nearly linearly with both the number of hyperedges and attributes. nodes (first cores, then fringes), where attachment probabil￾ities are governed by node attributes. We also introduced NOAHFIT, a parameter estimation algorithm that fits NOAH to a given hypergraph b… view at source ↗
read the original abstract

In many real-world scenarios, interactions happen in a group-wise manner with multiple entities, and therefore, hypergraphs are a suitable tool to accurately represent such interactions. Hyperedges in real-world hypergraphs are not composed of randomly selected nodes but are instead formed through structured processes. Consequently, various hypergraph generative models have been proposed to explore fundamental mechanisms underlying hyperedge formation. However, most existing hypergraph generative models do not account for node attributes, which can play a significant role in hyperedge formation. As a result, these models fail to reflect the interactions between structure and node attributes. To address the issue above, we propose NoAH, a stochastic hypergraph generative model for attributed hypergraphs. NoAH utilizes the core-fringe node hierarchy to model hyperedge formation as a series of node attachments and determines attachment probabilities based on node attributes. We further introduce NoAHFit, a parameter learning procedure that allows NoAH to replicate a given real-world hypergraph. Through experiments on nine datasets across four different domains, we show that NoAH with NoAHFit more accurately reproduces the structure-attribute interplay observed in the real-world hypergraphs than eight baseline hypergraph generative models, in terms of six metrics.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes NoAH, a stochastic generative model for attributed hypergraphs that represents hyperedge formation as a sequence of node attachments ordered by a core-fringe hierarchy, with attachment probabilities set exclusively by node attributes. It introduces NoAHFit, a parameter-learning procedure that tunes the model to reproduce a given real-world hypergraph, and reports that the fitted model outperforms eight baseline hypergraph generators on nine datasets using six metrics that quantify structure-attribute interplay.

Significance. If the central claims hold after addressing the fitting procedure, the work would supply a practical, interpretable mechanism for synthesizing attributed hypergraphs that preserve observed correlations between topology and node features. This could benefit downstream tasks such as link prediction, community detection, and simulation in social, biological, and collaboration networks. The broad experimental scope across four domains is a positive feature, though the reliance on post-hoc fitting limits the extent to which the results demonstrate a parameter-free generative principle.

major comments (2)
  1. [Abstract and §4] Abstract and §4 (NoAHFit): the claim that NoAH with NoAHFit 'more accurately reproduces the structure-attribute interplay' is achieved by construction, because NoAHFit explicitly optimizes attachment-probability parameters to match the target hypergraph. This circularity is load-bearing for the central claim; the reported superiority over attribute-ignorant baselines does not establish that the core-fringe, attribute-only generative process itself captures the interplay rather than merely fitting the observed statistics.
  2. [§3] §3 (Model definition): the generative process assumes that hyperedge formation is fully determined by attribute-driven attachment probabilities within a fixed core-fringe hierarchy. If real hyperedges arise from bidirectional structure-attribute effects (e.g., attribute homophily modulated by local density or existing hyperedges), the model will systematically underfit the interplay metrics even after NoAHFit; the manuscript does not provide a diagnostic test that distinguishes these cases.
minor comments (2)
  1. [Experimental evaluation] The experimental protocol should report the precise definitions and independence of the six metrics, together with any multiple-comparison corrections, so that the superiority claims can be evaluated quantitatively.
  2. [§5] Baseline implementations and hyperparameter choices for the eight comparison models should be documented in sufficient detail (or supplied as code) to permit exact reproduction of the reported metric values.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and insightful comments on our manuscript. We address the major comments point by point below, proposing clarifications and additions to the paper where the feedback identifies areas for improvement.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (NoAHFit): the claim that NoAH with NoAHFit 'more accurately reproduces the structure-attribute interplay' is achieved by construction, because NoAHFit explicitly optimizes attachment-probability parameters to match the target hypergraph. This circularity is load-bearing for the central claim; the reported superiority over attribute-ignorant baselines does not establish that the core-fringe, attribute-only generative process itself captures the interplay rather than merely fitting the observed statistics.

    Authors: We acknowledge that NoAHFit optimizes the attachment probability parameters to replicate the given real-world hypergraph, which includes matching the structure-attribute interplay metrics. This fitting procedure is indeed central to demonstrating the model's capability. However, the comparison with attribute-ignorant baselines is meaningful because those models lack any mechanism to incorporate node attributes, so they cannot reproduce attribute-related correlations regardless of fitting. Our results show that the specific generative process in NoAH—core-fringe hierarchy with attribute-driven attachments—allows for effective replication of the interplay when parameters are learned via NoAHFit. We will revise the abstract and §4 to more precisely describe NoAHFit as a parameter-learning method for replication and to frame the experimental results as evidence that this mechanism can capture the observed interplay, rather than implying a parameter-free generative principle. This revision will mitigate concerns about circularity. revision: partial

  2. Referee: [§3] §3 (Model definition): the generative process assumes that hyperedge formation is fully determined by attribute-driven attachment probabilities within a fixed core-fringe hierarchy. If real hyperedges arise from bidirectional structure-attribute effects (e.g., attribute homophily modulated by local density or existing hyperedges), the model will systematically underfit the interplay metrics even after NoAHFit; the manuscript does not provide a diagnostic test that distinguishes these cases.

    Authors: We agree with the referee that the model makes a strong assumption that hyperedge formation is determined by attribute-driven probabilities in a fixed core-fringe hierarchy, without explicit bidirectional feedback from structure to attributes. If real processes involve more complex interactions, such as density-modulated homophily, the model might not fully account for all dynamics. That said, the consistent outperformance on the six interplay metrics across nine datasets indicates that the proposed process provides a useful approximation. The manuscript does not include a diagnostic test to distinguish between generative assumptions, which is a limitation. In response, we will add text in §3 discussing the model's assumptions and potential limitations regarding bidirectional effects, and we will note the need for future diagnostic methods to compare alternative generative models. revision: yes

Circularity Check

1 steps flagged

NoAHFit parameter learning reduces reproduction of structure-attribute metrics to fitting by construction

specific steps
  1. fitted input called prediction [Abstract; NoAHFit description]
    "We further introduce NoAHFit, a parameter learning procedure that allows NoAH to replicate a given real-world hypergraph. Through experiments on nine datasets across four different domains, we show that NoAH with NoAHFit more accurately reproduces the structure-attribute interplay observed in the real-world hypergraphs than eight baseline hypergraph generative models, in terms of six metrics."

    NoAHFit learns parameters from the target hypergraph to make the model replicate it exactly. The subsequent claim of more accurately reproducing the structure-attribute interplay (via the six metrics) is therefore evaluated on the fitted replication of the same input data, reducing the 'prediction' or reproduction result to the fitting procedure by construction.

full rationale

The paper's central claim is that NoAH with NoAHFit reproduces the observed structure-attribute interplay more accurately than baselines on real hypergraphs. However, NoAHFit is explicitly a procedure to learn parameters so that the generative model replicates a given real-world hypergraph. The six metrics are then evaluated on this fitted output, making superior performance a direct consequence of the fitting step rather than an independent generative prediction. The core-fringe attachment process and attribute-based probabilities are inputs to this fit, so the reported match does not constitute a parameter-free derivation or out-of-sample prediction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim depends on the domain assumption that core-fringe hierarchy plus attribute-based attachment probabilities suffice to capture real hyperedge formation, plus parameters that are fitted to each target dataset.

free parameters (1)
  • attachment probability parameters
    Learned by NoAHFit to match a given real hypergraph
axioms (1)
  • domain assumption Hyperedge formation proceeds via successive node attachments in a core-fringe node hierarchy
    Invoked to structure the generative process

pith-pipeline@v0.9.0 · 5754 in / 1221 out tokens · 36028 ms · 2026-05-18T13:25:51.482556+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

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

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

43 extracted references · 43 canonical work pages

  1. [1]

    Hypergraph clustering by iteratively reweighted modularity maximization,

    T. Kumar, S. Vaidyanathan, H. Ananthapadmanabhan, S. Parthasarathy, and B. Ravindran, “Hypergraph clustering by iteratively reweighted modularity maximization,”Applied Network Science, vol. 5, no. 1, p. 52, 2020

    work page 2020

  2. [2]

    Hypergraph random walks, laplacians, and clustering,

    K. Hayashi, S. G. Aksoy, C. H. Park, and H. Park, “Hypergraph random walks, laplacians, and clustering,” inCIKM, 2020

    work page 2020

  3. [3]

    Adaptive hypergraph learning and its application in image classification,

    J. Yu, D. Tao, and M. Wang, “Adaptive hypergraph learning and its application in image classification,”TIP, vol. 21, no. 7, pp. 3262–3272, 2012

    work page 2012

  4. [4]

    Hypergraph-based anomaly detection of high- dimensional co-occurrences,

    J. Silva and R. Willett, “Hypergraph-based anomaly detection of high- dimensional co-occurrences,”TPAMI, vol. 31, no. 3, pp. 563–569, 2008

    work page 2008

  5. [5]

    Random walk with restart on hypergraphs: fast computation and an application to anomaly detection,

    J. Chun, G. Lee, K. Shin, and J. Jung, “Random walk with restart on hypergraphs: fast computation and an application to anomaly detection,” Data Mining and Knowledge Discovery, vol. 38, no. 3, pp. 1222–1257, 2024

    work page 2024

  6. [6]

    Structural patterns and generative models of real-world hypergraphs,

    M. T. Do, S. Yoon, B. Hooi, and K. Shin, “Structural patterns and generative models of real-world hypergraphs,” inKDD, 2020

    work page 2020

  7. [7]

    How do hyperedges overlap in real- world hypergraphs?-patterns, measures, and generators,

    G. Lee, M. Choe, and K. Shin, “How do hyperedges overlap in real- world hypergraphs?-patterns, measures, and generators,” inWWW, 2021

    work page 2021

  8. [8]

    How transitive are real-world group interactions?-measurement and reproduction,

    S. Kim, F. Bu, M. Choe, J. Yoo, and K. Shin, “How transitive are real-world group interactions?-measurement and reproduction,” inKDD, 2023

    work page 2023

  9. [9]

    Consistency of spectral hypergraph partitioning under planted partition model,

    D. Ghoshdastidar and A. Dukkipati, “Consistency of spectral hypergraph partitioning under planted partition model,”The Annals of Statistics, vol. 45, no. 1, pp. 289–315, 2017

    work page 2017

  10. [10]

    Community detection in large hypergraphs,

    N. Ruggeri, M. Contisciani, F. Battiston, and C. De Bacco, “Community detection in large hypergraphs,”Science Advances, vol. 9, no. 28, p. eadg9159, 2023

    work page 2023

  11. [11]

    Structure and inference in hypergraphs with node attributes,

    A. Badalyan, N. Ruggeri, and C. De Bacco, “Structure and inference in hypergraphs with node attributes,”Nature Communications, vol. 15, no. 1, p. 7073, 2024

    work page 2024

  12. [12]

    Inference of hyperedges and overlapping communities in hypergraphs,

    M. Contisciani, F. Battiston, and C. De Bacco, “Inference of hyperedges and overlapping communities in hypergraphs,”Nature Communications, vol. 13, no. 1, p. 7229, 2022

    work page 2022

  13. [13]

    Birds of a feather: Homophily in social networks,

    M. McPherson, L. Smith-Lovin, and J. M. Cook, “Birds of a feather: Homophily in social networks,”Annual Review of Sociology, vol. 27, no. 1, pp. 415–444, 2001

    work page 2001

  14. [14]

    Villain: Self-supervised learning on ho- mogeneous hypergraphs without features via virtual label propagation,

    G. Lee, S. Y . Lee, and K. Shin, “Villain: Self-supervised learning on ho- mogeneous hypergraphs without features via virtual label propagation,” inWWW, 2024

    work page 2024

  15. [15]

    Graph mining: Laws, generators, and algorithms,

    D. Chakrabarti and C. Faloutsos, “Graph mining: Laws, generators, and algorithms,”ACM Computing Surveys, vol. 38, no. 1, pp. 2–es, 2006

    work page 2006

  16. [16]

    A survey on hypergraph mining: Patterns, tools, and generators,

    G. Lee, F. Bu, T. Eliassi-Rad, and K. Shin, “A survey on hypergraph mining: Patterns, tools, and generators,”ACM Computing Surveys, vol. 57, no. 8, pp. 1–36, 2025

    work page 2025

  17. [17]

    Random spatial network models for core- periphery structure,

    J. Jia and A. R. Benson, “Random spatial network models for core- periphery structure,” inWSDM, 2019

    work page 2019

  18. [18]

    Core-periphery models for hyper- graphs,

    M. Papachristou and J. Kleinberg, “Core-periphery models for hyper- graphs,” inKDD, 2022

    work page 2022

  19. [19]

    Growth patterns and models of real-world hypergraphs,

    J. Ko, Y . Kook, and K. Shin, “Growth patterns and models of real-world hypergraphs,”Knowledge and Information Systems, vol. 64, no. 11, pp. 2883–2920, 2022

    work page 2022

  20. [20]

    Network structure and minimum degree,

    S. B. Seidman, “Network structure and minimum degree,”Social Net- works, vol. 5, no. 3, pp. 269–287, 1983

    work page 1983

  21. [21]

    Models of core/periphery structures,

    S. P. Borgatti and M. G. Everett, “Models of core/periphery structures,” Social Networks, vol. 21, no. 4, pp. 375–395, 2000

    work page 2000

  22. [22]

    Hypercore decomposition for non-fragile hyperedges: concepts, algorithms, observations, and applications,

    F. Bu, G. Lee, and K. Shin, “Hypercore decomposition for non-fragile hyperedges: concepts, algorithms, observations, and applications,”Data Mining and Knowledge Discovery, vol. 37, no. 6, p. 2389–2437, 2023

    work page 2023

  23. [23]

    Core-periphery detection in hypergraphs,

    F. Tudisco and D. J. Higham, “Core-periphery detection in hypergraphs,” SIAM Journal on Mathematics of Data Science, vol. 5, no. 1, pp. 1–21, 2023

    work page 2023

  24. [24]

    Graphvae: Towards generation of small graphs using variational autoencoders,

    M. Simonovsky and N. Komodakis, “Graphvae: Towards generation of small graphs using variational autoencoders,” inICANN, 2018

    work page 2018

  25. [25]

    Netgan: Generating graphs via random walks,

    A. Bojchevski, O. Shchur, D. Z ¨ugner, and S. G ¨unnemann, “Netgan: Generating graphs via random walks,” inICML, 2018

    work page 2018

  26. [26]

    De Cao and T

    N. De Cao and T. Kipf, “Molgan: An implicit generative model for small molecular graphs,”arXiv preprint arXiv:1805.11973, 2018

    work page arXiv 2018

  27. [27]

    Generative diffusion models on graphs: Methods and applications,

    C. Liu, W. Fan, Y . Liu, J. Li, H. Li, H. Liu, J. Tang, and Q. Li, “Generative diffusion models on graphs: Methods and applications,” in IJCAI, 2023

    work page 2023

  28. [28]

    An introduction to exponential random graph (p*) models for social networks,

    G. Robins, P. Pattison, Y . Kalish, and D. Lusher, “An introduction to exponential random graph (p*) models for social networks,”Social Networks, vol. 29, no. 2, pp. 173–191, 2007

    work page 2007

  29. [29]

    Mixed membership stochastic blockmodels,

    E. M. Airoldi, D. Blei, S. Fienberg, and E. Xing, “Mixed membership stochastic blockmodels,”NeurIPS, 2008

    work page 2008

  30. [30]

    Joint latent space model for social networks with multivariate attributes,

    S. Wang, S. Paul, and P. De Boeck, “Joint latent space model for social networks with multivariate attributes,”Psychometrika, vol. 88, no. 4, pp. 1197–1227, 2023

    work page 2023

  31. [31]

    Multiplicative attribute graph model of real- world networks,

    M. Kim and J. Leskovec, “Multiplicative attribute graph model of real- world networks,”Internet Mathematics, vol. 8, no. 1-2, pp. 113–160, 2012

    work page 2012

  32. [32]

    Planted hitting set recovery in hypergraphs,

    I. Amburg, J. Kleinberg, and A. R. Benson, “Planted hitting set recovery in hypergraphs,”Journal of Physics: Complexity, vol. 2, no. 3, p. 035004, 2021

    work page 2021

  33. [33]

    Combinatorial characteriza- tions and impossibilities for higher-order homophily,

    N. Veldt, A. R. Benson, and J. Kleinberg, “Combinatorial characteriza- tions and impossibilities for higher-order homophily,”Science Advances, vol. 9, no. 1, p. eabq3200, 2023

    work page 2023

  34. [34]

    When hypergraph meets heterophily: New benchmark datasets and baseline,

    M. Li, Y . Gu, Y . Wang, Y . Fang, L. Bai, X. Zhuang, and P. Lio, “When hypergraph meets heterophily: New benchmark datasets and baseline,” inAAAI, 2025

    work page 2025

  35. [35]

    Attributed hypergraph generation with realistic interplay between structure and attributes: Code, datasets, and online appendix,

    J. Chun, S. Yoon, M. Choe, G. Lee, and K. Shin, “Attributed hypergraph generation with realistic interplay between structure and attributes: Code, datasets, and online appendix,” 2025. [Online]. Available: https://github.com/jaewan01/NoAH

    work page 2025

  36. [36]

    Hypergcn: A new method for training graph convolutional networks on hypergraphs,

    N. Yadati, M. Nimishakavi, P. Yadav, V . Nitin, A. Louis, and P. Talukdar, “Hypergcn: A new method for training graph convolutional networks on hypergraphs,”NeurIPS, 2019

    work page 2019

  37. [37]

    Generative hypergraph clustering: From blockmodels to modularity,

    P. S. Chodrow, N. Veldt, and A. R. Benson, “Generative hypergraph clustering: From blockmodels to modularity,”Science Advances, vol. 7, no. 28, p. eabh1303, 2021

    work page 2021

  38. [38]

    Can co-location be used as a proxy for face- to-face contacts?

    M. G ´enois and A. Barrat, “Can co-location be used as a proxy for face- to-face contacts?”EPJ Data Science, vol. 7, no. 1, pp. 1–18, 2018

    work page 2018

  39. [39]

    Justifying recommendations using distantly-labeled reviews and fine-grained aspects,

    J. Ni, J. Li, and J. McAuley, “Justifying recommendations using distantly-labeled reviews and fine-grained aspects,” inEMNLP, 2019

    work page 2019

  40. [40]

    Amburg, N

    I. Amburg, N. Veldt, and A. R. Benson, “Fair clustering for diverse and experienced groups,”arXiv:2006.05645, 2020

    work page arXiv 2006

  41. [41]

    Evolution of real-world hypergraphs: Patterns and models without oracles,

    Y . Kook, J. Ko, and K. Shin, “Evolution of real-world hypergraphs: Patterns and models without oracles,” inICDM, 2020

    work page 2020

  42. [42]

    Randomizing hypergraphs preserving degree correlation and local clustering,

    K. Nakajima, K. Shudo, and N. Masuda, “Randomizing hypergraphs preserving degree correlation and local clustering,”IEEE TNSE, vol. 9, no. 3, pp. 1139–1153, 2021

    work page 2021

  43. [43]

    Kronecker generative models for power-law patterns in real-world hypergraphs,

    M. Choe, J. Ko, T. Kwon, K. Shin, and C. Faloutsos, “Kronecker generative models for power-law patterns in real-world hypergraphs,” inWWW, 2025

    work page 2025