Pith. sign in

REVIEW 4 major objections 5 minor 77 references

Beyond Connectivity: Higher-Order Network Framework for Capturing Memory-Driven Mobility Dynamics

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that modeling mobility with a third-order memory—where each network state is a subpath of the last three intersections—outperforms memoryless first-order models on centrality and prediction, selecting $k=3$ by…

desk verdict A competent application of existing higher-order network methods to transportation, but the headline out-of-sample claims are not yet supported by the evaluation. read the letter →

arxiv 2507.07727 v1 pith:TDCRFLL4 submitted 2025-07-10 cs.SI

Chen Zhang , Jürgen Hackl This is my paper
classification cs.SI MSC 05C8260J2090B20
keywords higher-ordernetworksmemoryeffectstransportationdeBruijngraphsMarkovchainsbetweennesscentralityPageRanknext-stepprediction
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

The paper tries to establish that urban mobility cannot be treated as a memoryless random walk: where a traveler has been in the last few steps changes where they go next, and network analyses that ignore this are systematically worse. It builds higher-order networks in which a node is a subpath of $k$ consecutive intersections rather than a single intersection, with edges weighted by transition probabilities learned from observed trajectories. On agent-based trajectory data for the Sioux Falls road network, the resulting models beat first-order baselines on betweenness centrality, PageRank, and next-step prediction, with a third-order model (memory of the last three steps) giving the best balance of accuracy and complexity. If this holds, traffic forecasting, bottleneck identification, and resilience analysis should incorporate memory effects rather than connectivity alone.

What carries the argument

The de Bruijn graph, or subpath network, of order $k$: each node is a sequence of $k$ consecutive intersections $(v_1, \ldots, v_k)$, and a directed edge connects $(v_1, \ldots, v_k)$ to $(v_2, \ldots, v_{k+1})$ with weight equal to the empirical conditional probability $P(v_{k+1} \mid v_k, \ldots, v_1)$. This machinery turns a trajectory dataset into a memory-aware Markov chain, lets every classical measure be re-run on states that encode history, and provides the likelihood-ratio framework that picks the memory length. It is what allows the same network toolbox (shortest paths, stationary distributions, transition probabilities) to be applied without assuming forgetting.

What would settle it

Run the same model-selection and prediction procedure on real trajectory data from a comparable city, such as GPS traces or smart-card records with full trip sequences. If a first-order model matches or beats the third-order model on held-out next-step accuracy, or if the likelihood-ratio test picks $k=1$, the central claim fails as stated. A cheaper check is to compare the empirical conditional probabilities $P(v_{t+1} \mid v_t)$ versus $P(v_{t+1} \mid v_t, v_{t-1})$ in the MATSim data with those in real traces; if real traces show no additional predictability from the second step, the reported memory effect is an artifact of the simulator.

Watch

Extended reading notes

Core claim

The central claim is that a $k$th-order de Bruijn-style network, where each state is a $k$-node path and each edge is a transition between overlapping subpaths, captures the sequential structure of travel better than the original graph. Generalized betweenness centrality counts shortest paths in this subpath network and attributes them to original nodes; generalized PageRank is the stationary distribution of a random walk on subpath states, aggregated to the last component; next-step prediction uses the conditional probability $P(v_i \mid v_{i-1}, \ldots, v_{i-k})$. Estimated by maximum likelihood from path frequencies, these measures align more closely with ground-truth node visitation and traversal frequencies: Kendall's tau for betweenness rises from $0.185$ to $0.452$ and KL divergence falls from $1.597$ to $0.034$ at the optimal order; prediction accuracy rises from $62.1\%$ to $87.5\%$. A likelihood-ratio test over nested multi-order models selects $k^*=3$ as optimal, with higher orders overfitting and lower orders underfitting.

Load-bearing premise

The whole comparison rests on the assumption that MATSim-generated trajectories on the enriched Sioux Falls network reproduce the sequential dependencies of real urban travel, so the gains measured on synthetic data would also show up on real trajectories.

Editorial extensions

If this is right

  • Centrality rankings from memory-aware models match observed flow better, so bottleneck and congestion analyses based on first-order betweenness may miss the road segments that actually carry traffic.
  • Next-step prediction on these data improves from $62.1\%$ accuracy with a first-order model to $87.5\%$ at third order, with cross-entropy loss dropping from $0.728$ to $0.313$, so route and flow forecasts are materially more accurate.
  • The likelihood-ratio selection of $k^*=3$ means three preceding steps is a sufficient statistic for routing behavior on this network; orders above three add complexity without predictive gain.
  • Because the method also works in a non-attributed form (connectivity only), memory effects can be captured even where trajectory frequencies are unavailable, only the sequence of visited nodes.
  • The same construction applies beyond Sioux Falls, to GPS traces, phone records, or smart-card data, whenever paths are observed as sequences.

Reading between the lines

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

  • A natural testable extension is to repeat the pipeline on real GPS or smart-card trajectories; if first-order wins there, the reported gains would be an artifact of the simulator's route-choice model rather than a property of human mobility.
  • The optimal order likely depends on network size, trip-length distribution, and route-choice heterogeneity; $k=3$ should not be read as a universal constant even if the mechanism is real.
  • The same subpath-state representation could feed graph-learning models, letting a neural net learn variable memory depths per region instead of a single fixed order, which the paper lists as future work but does not itself test.
  • Because attributed first-order betweenness had worse KL divergence than non-attributed, the results suggest that naively adding empirical weights to a memoryless model can mislead; this is worth checking as a separate hypothesis.
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 / 5 minor

Summary. The paper introduces a higher-order network framework for modeling memory-dependent mobility, in which kth-order de Bruijn graph nodes represent traversed subpaths and edge weights encode empirical transition probabilities. The authors generalize betweenness centrality, PageRank, and next-step prediction to this representation and evaluate the framework on MATSim-generated trajectories over an enriched Sioux Falls network. They report that a third-order model, selected by likelihood-ratio tests, outperforms first-order baselines across all three tasks (Tables 2-4), and they discuss scalability, limitations, and future directions.

Significance. If the reported results hold, the paper would provide a useful application-level demonstration that memory effects matter in transportation network analytics and prediction. The manuscript is clearly written, uses established open-source tools (MATSim, pathpyG, NetworkX, cuGraph), shares code via GitHub, and includes a formal, if elementary, unbiasedness proposition. However, the empirical support for the central claim is currently weakened by in-sample centrality evaluation and by an unspecified split protocol for the prediction task; the significance of the work therefore depends on whether the evaluation can be made genuinely out-of-sample.

major comments (4)
  1. [Section 4.2 and Tables 2-3] The centrality evaluations are in-sample. For both betweenness centrality and PageRank, the ground truth is the node (or subpath) frequency computed from the same trajectories that are used to estimate the empirical transition probabilities in Eq. (3) and, for attributed models, the edge weights. Comparing model output with this ground truth measures goodness-of-fit on the estimation data, not predictive or structural accuracy. Because higher-order models contain more parameters, some improvement is expected by construction. The authors should evaluate centrality on held-out trajectories (e.g., fit on a training set and compare against test-trajectory frequencies) or otherwise decouple the ground truth from the model inputs.
  2. [Section 4.2, Table 4, and Figure 9] The next-step prediction split is not fully specified. The paper states that 983,110 subpaths are split 50/50 into training and test sets, but does not state whether the split is performed at the level of trajectories or at the level of individual overlapping windows. If overlapping windows from the same trajectory appear in both folds, the test set contains the exact context needed for prediction, which would inflate the reported accuracy gains (62.1% to 87.5%). The authors must clarify that the split is by trajectory and report results under that condition.
  3. [Section 5 (Optimal-Order paragraph) and Tables 2-4] The claim that the third-order model is globally optimal is not fully supported by the reported metrics. The likelihood-ratio test selects k*=3 on the full data, and the subsequent evaluation is performed on the same data, so there is no independent validation of the selected order. Moreover, Table 4 shows that fifth-order accuracy (88.5%) exceeds third-order accuracy (87.5%), and Table 3 shows that PageRank KL divergence decreases monotonically through k=5. The 'optimal balance' claim is therefore metric-dependent and should be qualified accordingly, or the model-selection and evaluation should be performed on separate data.
  4. [Tables 2-5] All results are reported as single point estimates without error bars, confidence intervals, or repeated runs, despite the stochastic nature of the MATSim simulation. Given that several comparisons involve small differences (e.g., Kendall's tau of 0.452 vs. 0.440 at orders 3 and 4 in Table 2), the reader cannot assess whether these differences are statistically meaningful. The authors should provide uncertainty quantification, at minimum by repeating the simulation or bootstrapping the trajectory sample.
minor comments (5)
  1. [Throughout] There are several typographical errors, including 'T ransportation' in the running title, 'F ramework' in the header, 'sums oval' in the description of Eq. (3), 'Los [-)' in Table 4, and 'reaching 88.%' in the text near Table 4. A careful proofreading pass is needed.
  2. [Section 2.2, Proposition 1] The unbiasedness proof conditions on Y but does not address the case Y=0, and the binomial assumption X|Y would only hold if the true data-generating order is k. The proposition should be stated with the necessary support conditions and a note that misspecification of the order can introduce bias.
  3. [Section 6, Discussion] The paragraph beginning 'In real-world applications where detailed trajectory data is not available...' contains a duplicated sentence about the fully data-driven approach; the duplicate should be removed.
  4. [Introduction and Appendix C] The introduction promises a 'dual-scale analysis' but this term is not defined or used later. The multi-order model in Appendix C is described as applying lower-order models at the start of a path, but the main text's next-step prediction section does not clearly connect this formulation to the experimental implementation.
  5. [References] Reference [45] contains a typo ('Survay'), and the GitHub link in Section 5 states that code will be archived on Zenodo only upon acceptance; for reproducibility, a permanent DOI should be provided at revision time.

Circularity Check

2 steps flagged · score 6.0 of 10

Centrality and PageRank evaluations are in-sample fits to the same empirical frequencies; only the possibly leaked held-out next-step split provides out-of-sample evidence.

  1. fitted input called prediction [Section 4.2 (Experimental Setup), with Eq. (3) and Eqs. (8)-(9)]
    "For PageRank, ground truth is based on node visitation frequencies aggregated over all simulated agent trajectories. ... In contrast, the PageRank and next-step prediction analyses directly use the conditional transition probabilities as edge weights, reflecting the empirically derived likelihood of movement between nodes or subpaths."

    The transition probabilities are the MLE empirical conditional frequencies estimated from the same MATSim trajectories that define the ground-truth node visitation frequencies. For a Markov chain estimated from a trajectory corpus, the stationary distribution of the empirical transition matrix is, up to boundary effects, the empirical visitation-frequency vector itself; with the damping factor it is a smoothed version of the same vector. Hence the 'predicted' PageRank is a deterministic function of the ground-truth frequencies, and the reported KL/Kendall improvements with order measure how well the empirical transition matrix reproduces the very frequencies used to estimate it. This is fitted input called prediction, not an out-of-sample structural validation.

  2. fitted input called prediction [Section 4.2 (Experimental Setup), Section 3.1 Eq. (7), Table 2]
    "For the betweenness centrality analysis, edge weights are defined as the negative logarithm of conditional transition probabilities, assigning lower costs to more frequently traversed paths. ... For betweenness centrality, the ground truth corresponds to the frequency with which a node v (in first-order models) or a subpath v(k) (in higher-order models) is traversed along paths between origin-destination pairs."

    The cost function for betweenness centrality is the negative logarithm of the same empirical conditional transition probabilities inferred from the trajectory corpus whose subpath/traversal frequencies serve as ground truth. Shortest paths in the kth-order network therefore are, by construction, the most probable paths under the empirical counts, so the centrality ranking is a transformed version of those counts. Comparing this ranking to the ground-truth traversal frequencies is an in-sample goodness-of-fit exercise; higher-order models possess more parameters and are expected to fit the training corpus better. The reported superiority in Table 2 thus does not demonstrate that the models 'predict' structurally critical components beyond the data used to weight the network.

full rationale

The mathematical construction of higher-order Markov models from empirical subpath frequencies (Eq. 3) is standard maximum likelihood and is not itself circular; the de Bruijn/line-graph representation is a legitimate modeling choice, and I found no load-bearing self-citation or imported uniqueness theorem. The circularity is confined to the evaluation protocol for two of the three headline results. For PageRank, the model uses the empirical conditional transition probabilities as edge weights, while the ground truth is the node visitation frequency of the very same MATSim trajectories; for a Markov chain estimated from a trajectory corpus, the stationary distribution of the empirical transition matrix is (up to boundary effects) the empirical visitation distribution, so the reported KL/Kendall agreement is an in-sample fit of a smoothed version of the target. For betweenness centrality, the edge costs are negative logarithms of the same empirical transition probabilities and the ground truth is the traversal frequency of the same subpaths, so shortest-path centrality again is a transformed version of the training counts; higher-order models have more parameters and are expected to fit better in-sample. The only task with a held-out component, next-step prediction, is reported as a 50/50 split of 983,110 subpaths, but the split unit is unspecified; if subpaths are split randomly, overlapping sliding windows from the same trajectory leak the conditioning context into both train and test, so the large accuracy gain is not conclusively out-of-sample. The paper also acknowledges that the synthetic MATSim data 'may not fully capture the diversity of real-world travel behavior' (Section 6), which is an external-validity limitation rather than a circularity. Overall, the core methodology is not circular, but two of the three 'predictions' reduce to in-sample fits of the same empirical frequencies, yielding a partial circularity score of 6.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The framework itself introduces no new free parameters or entities beyond standard higher-order network machinery. The evaluation relies on several human choices (alpha, epsilon, max order) and on a domain assumption that synthetic MATSim trajectories are representative. The main circular element is the use of the same data for both fitting and evaluation of centrality.

free parameters (4)
  • PageRank damping factor alpha = not reported (likely default 0.85)
    Used in Eq. 8 for all PageRank experiments; the paper states alpha in (0,1) but never reports the chosen value.
  • LRT significance threshold epsilon = 0.05
    Used in Appendix C to select optimal order; chosen by convention, not derived from the data.
  • maximum order K = 5
    The paper evaluates orders 1 through 5 and calls order 3 optimal; the choice to stop at 5 is not justified by data.
  • optimal order k* = 3
    Selected by consecutive likelihood-ratio tests on the same trajectory data (Section 5); the selection is part of the claimed result, not an independent input.
assumptions (4)
  • domain assumption Trajectories are statistically independent and identically distributed samples of a kth-order Markov process
    Assumed in Proposition 1 and in the likelihood-ratio test for order selection (Section 2.2, Appendix C); real mobility trajectories are rarely i.i.d. due to correlations across agents and time.
  • domain assumption The MATSim-generated synthetic trajectories capture real-world memory effects in routing
    The entire validation rests on this; the paper acknowledges synthetic data 'may not fully capture the diversity of real-world travel behavior' (Section 6).
  • ad hoc to paper Ground truth for centrality is the empirical node frequency in the observed paths
    This choice makes the evaluation in-sample and privileges models that memorize the data (Section 4.2).
  • standard math Wilks' theorem applies to the likelihood-ratio statistic for model order selection
    Invoked in Appendix C to justify a chi-square distribution; requires nested models and regularity conditions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Connectivity: Higher-Order Network Framework for Capturing Memory-Driven Mobility Dynamics." pith.science (2026). https://pith.science/paper/TDCRFLL4

@misc{pith2026250707727,
  author       = {Pith},
  title        = {Pith review of: Beyond Connectivity: Higher-Order Network Framework for Capturing Memory-Driven Mobility Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TDCRFLL4}},
  note         = {Machine review of arXiv:2507.07727}
}
read the original abstract

Understanding and predicting mobility dynamics in transportation networks is critical for infrastructure planning, resilience analysis, and traffic management. Traditional graph-based models typically assume memoryless movement, limiting their ability to capture sequential dependencies inherent in real-world mobility patterns. In this study, we introduce a novel higher-order network framework for modeling memory-dependent dynamics in transportation systems. By extending classical graph representations through higher-order Markov chains and de Bruijn graph structures, our framework encodes the spatial and temporal ordering of traversed paths, enabling the analysis of structurally and functionally critical components with improved fidelity. We generalize key network analytics, including betweenness centrality, PageRank, and next-step prediction, to this higher-order setting and validate our approach on the Sioux Falls transportation network using agent-based trajectory data generated with MATSim. Experimental results demonstrate that higher-order models outperform first-order baselines across multiple tasks, with the third-order model achieving an optimal balance between predictive accuracy and model complexity. These findings highlight the importance of incorporating memory effects into network-based transportation analysis and offer a scalable, data-driven methodology for capturing complex mobility behaviors in infrastructure systems.

Figures

Figures reproduced from arXiv: 2507.07727 by the authors.

Figure 1
Figure 1. Hierarchical, bottom-up construction of a Higher-Order Network (HON). At [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Aggregation Multi-Order Network that visualizes a random walk process [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Overview of the application example. (a) Visualization of the Sioux Falls road [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Experimental Workflow. A schematic overview of the modeling pipeline [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: Comparison of betweenness centrality measures across different network [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]
Figure 6
Figure 6. Figure 6: Kernel Density Estimation (KDE) and frequency distribution plots of Be [PITH_FULL_IMAGE:figures/full_fig_p020_6.png]
Figure 7
Figure 7. Figure 7: Comparison of network centrality measures across different network orders [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]
Figure 8
Figure 8. Figure 8: Kernel Density Estimation (KDE) and frequency distribution plots of PageR [PITH_FULL_IMAGE:figures/full_fig_p022_8.png]
Figure 9
Figure 9. Figure 9: Next-step prediction accuracy for the Sioux Falls network using higher-order [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

77 extracted references · 49 canonical work pages

  1. [1]

    L., Tortosa, L., and Vicent, J

    Agryzkov, T., Oliver, J. L., Tortosa, L., and Vicent, J. F., An algorithm for rank- ing the nodes of an urban network based on the concept of PageRank vector, Ap- plied Mathematics and Computation 219 (2012) 2186–2193, doi:10.1016/j.amc.2012. 08.064

    work page doi:10.1016/j.amc.2012 2012

  2. [2]

    and Estrada, E., Communicability geometry captures traffic flows in cities, Nature Human Behaviour 2 (2018) 645–652, doi:10.1038/s41562-018-0407-3

    Akbarzadeh, M. and Estrada, E., Communicability geometry captures traffic flows in cities, Nature Human Behaviour 2 (2018) 645–652, doi:10.1038/s41562-018-0407-3

    doi:10.1038/s41562-018-0407-3 2018

  3. [3]

    Aleta, A., Meloni, S., and Moreno, Y., A Multilayer perspective for the analysis of urban transportation systems, Scientific Reports 7 (2017) 44359, doi:10.1038/ srep44359

    work page 2017

  4. [4]

    (Plume, New York, NY, 2009)

    Barab´ asi, A.-L.,Linked: How Everything Is Connected to Everything Else and What It Means for Business, Science, and Everyday Life , A Plume Book, 20th edn. (Plume, New York, NY, 2009)

    work page 2009

  5. [5]

    and Albert, R., Emergence of Scaling in Random Networks, Science 286 (1999) 509–512, doi:10.1126/science.286.5439.509

    Barab´ asi, A.-L. and Albert, R., Emergence of Scaling in Random Networks, Science 286 (1999) 509–512, doi:10.1126/science.286.5439.509

    doi:10.1126/science.286.5439.509 1999

  6. [6]

    and Bonabeau, E., Scale-Free Networks, Scientific American 288 (2003) 60–69, doi:10.1038/scientificamerican0503-60

    Barab´ asi, A.-L. and Bonabeau, E., Scale-Free Networks, Scientific American 288 (2003) 60–69, doi:10.1038/scientificamerican0503-60

    work page doi:10.1038/scientificamerican0503-60 2003

  7. [7]

    M., Peixoto, T

    Battiston, F., Amico, E., Barrat, A., Bianconi, G., Ferraz de Arruda, G., Franceschiello, B., Iacopini, I., K´ efi, S., Latora, V., Moreno, Y., Murray, M. M., Peixoto, T. P., Vaccarino, F., and Petri, G., The physics of higher-order inter- actions in complex systems, Nature Physics 17 (2021) 1093–1098, doi:10.1038/ s41567-021-01371-4

    work page 2021

  8. [8]

    Bavelas, A., Communication Patterns in Task-Oriented Groups, The Journal of the Acoustical Society of America 22 (1950) 725–730, doi:10.1121/1.1906679

    doi:10.1121/1.1906679 1950

Show all 77 references
  1. [9]

    R., Gleich, D

    Benson, A. R., Gleich, D. F., and Leskovec, J., Higher-order organization of complex networks, Science 353 (2016) 163–166, doi:10.1126/science.aad9029

    2016 doi

  2. [10]

    Besenczi, R., B´ atfai, N., Jeszenszky, P., Major, R., Monori, F., and Isp´ any, M., Large- scale simulation of traffic flow using Markov model, PLOS ONE 16 (2021) e0246062, doi:10.1371/journal.pone.0246062

    2021 doi

  3. [11]

    G., de, A combinatorial problem, Proceedings of the Koninklijke Neder- landse Akademie van Wetenschappen 49 (1946) 758–764

    Bruijn, N. G., de, A combinatorial problem, Proceedings of the Koninklijke Neder- landse Akademie van Wetenschappen 49 (1946) 758–764

    1946

  4. [12]

    Cascetta, E., A stochastic process approach to the analysis of temporal dynamics in transportation networks, Transportation Research Part B: Methodological 23 (1989) 1–17, doi:10.1016/0191-2615(89)90019-2

    1989 doi

  5. [13]

    and Fourie, P

    Chakirov, A. and Fourie, P. J., Enriched Sioux Falls scenario with dynamic and disaggregate demand (2014) 39 p., doi:10.3929/ETHZ-B-000080996

    2014 doi

  6. [14]

    Chang, S., Pierson, E., Koh, P. W., Gerardin, J., Redbird, B., Grusky, D., and Leskovec, J., Mobility network models of COVID-19 explain inequities and inform reopening, Nature 589 (2021) 82–87, doi:10.1038/s41586-020-2923-3

    2021 doi

  7. [15]

    Compeau, P. E. C., Pevzner, P. A., and Tesler, G., How to apply de Bruijn graphs to genome assembly, Nature Biotechnology 29 (2011) 987–991, doi:10.1038/nbt.2023

    2011 doi

  8. [16]

    Crucitti, P., Latora, V., and Porta, S., Centrality in networks of urban streets, Chaos: An Interdisciplinary Journal of Nonlinear Science 16 (2006) 015113, doi:10.1063/1. 2150162

    2006 doi

  9. [17]

    Eliassi-Rad, T., Latora, V., Rosvall, M., and Scholtes, I., Higher-Order Graph Models: From Theoretical Foundations to Machine Learning,Dagstuhl Reports 11 (2021) 154, doi:10.4230/DagRep.11.7.139

    2021 doi

  10. [18]

    Y., Duarte, F., Qiu, W., Hackl, J., and Skibniewski, M

    Fan, X., Jiang, Y., Zhong, R. Y., Duarte, F., Qiu, W., Hackl, J., and Skibniewski, M. J., Advancements and potential pitfalls for smart cities and sentient infrastruc- tures, Digital Engineering 5 (2025) 100036, doi:10.1016/j.dte.2025.100036. September 3, 2025 17:43 manuscript...

    2025

  11. [19]

    C., A Set of Measures of Centrality Based on Betweenness, Sociometry 40 (1977) 35–41, doi:10.2307/3033543

    Freeman, L. C., A Set of Measures of Centrality Based on Betweenness, Sociometry 40 (1977) 35–41, doi:10.2307/3033543

    1977 doi

  12. [20]

    and Marble, D., The structure of transportation networks (1962) 73–88

    Garrison, W. and Marble, D., The structure of transportation networks (1962) 73–88

    1962

  13. [21]

    L., CONNECTIVITY OF THE INTERSTATE HIGHW AY SYSTEM, Papers in Regional Science 6 (1960) 121–137, doi:10.1111/j.1435-5597.1960.tb01707

    Garrison, W. L., CONNECTIVITY OF THE INTERSTATE HIGHW AY SYSTEM, Papers in Regional Science 6 (1960) 121–137, doi:10.1111/j.1435-5597.1960.tb01707. x

    1960

  14. [22]

    C., Hidalgo, C

    Gonz´ alez, M. C., Hidalgo, C. A., and Barab´ asi, A.-L., Understanding individual hu- man mobility patterns, Nature 453 (2008) 779–782, doi:10.1038/nature06958

    2008 doi

  15. [23]

    and Adey, B

    Hackl, J. and Adey, B. T., Generation of Spatially Embedded Random Networks to Model Complex Transportation Networks, in 14th International Probabilistic Work- shop, eds. Caspeele, R., Taerwe, L., and Proske, D. (Springer International Publishing, Cham, 2017), ISBN 978-3-319-4...

    2017

  16. [24]

    and Adey, B

    Hackl, J. and Adey, B. T., Estimation of traffic flow changes using networks in networks approaches, Applied Network Science 4 (2019) 28, doi:10.1007/ s41109-019-0139-y

    2019

  17. [25]

    and Adey, B

    Hackl, J. and Adey, B. T., Modelling multi-layer spatially embedded random net- works, Journal of Complex Networks 7 (2019) 254–280, doi:10.1093/comnet/cny019

    2019 doi

  18. [26]

    Hackl, J., Adey, B. T., and Lethanh, N., Determination of Near-Optimal Restoration Programs for Transportation Networks Following Natural Hazard Events Using Simu- lated Annealing: Restoration programs for transportation networks following natural hazard events using simulated...

    2018 doi

  19. [27]

    and Dubernet, T., Epidemic Spreading in Urban Areas Using Agent-Based Transportation Models, Future Internet 11 (2019) 92, doi:10.3390/fi11040092

    Hackl, J. and Dubernet, T., Epidemic Spreading in Urban Areas Using Agent-Based Transportation Models, Future Internet 11 (2019) 92, doi:10.3390/fi11040092

    2019 doi

  20. [28]

    Hackl, J., Scholtes, I., Petrovi´ c, L. V., Perri, V., Verginer, L., and Gote, C., Anal- ysis and visualisation of time series data on networks with pathpy, in Companion Proceedings of the Web Conference , WWW ’21 (Association for Computing Ma- chinery, New York, NY, USA, 2021...

    2021

  21. [29]

    A., Schult, D

    Hagberg, A. A., Schult, D. A., and Swart, P. J., Exploring Network Structure, Dy- namics, and Function using NetworkX, in Python in Science Conference (Pasadena, California, 2008), pp. 11–15, doi:10.25080/TCWV9851

    2008 doi

  22. [30]

    and Chorley, R., Network Analysis in Geography, Explorations in Spatial Structure (Edward Arnold, 1969)

    Haggett, P. and Chorley, R., Network Analysis in Geography, Explorations in Spatial Structure (Edward Arnold, 1969)

    1969

  23. [31]

    and and Bliemer, M

    Haghani, M. and and Bliemer, M. C. J., Emerging trends and influential outsiders of transportation science, Transportation Letters 15 (2023) 386–422, doi:10.1080/ 19427867.2022.2057397

    2023

  24. [32]

    and Scholtes, I., Using Causality-Aware Graph Neural Networks to Predict Temporal Centralities in Dynamic Graphs (2023)

    Heeg, F. and Scholtes, I., Using Causality-Aware Graph Neural Networks to Predict Temporal Centralities in Dynamic Graphs (2023)

    2023

  25. [33]

    W., The Multi-Agent Transport Simulation MATSim (Ubiquity Press, 2016)

    Horni, A., Nagel, K., and Axhausen, K. W., The Multi-Agent Transport Simulation MATSim (Ubiquity Press, 2016)

    2016

  26. [34]

    Jiang, B., Ranking spaces for predicting human movement in an urban environment, International Journal of Geographical Information Science 23 (2009) 823–837, doi: 10.1080/13658810802022822

    2009 doi

  27. [35]

    84 in University of Chicago

    Kansky, K., Structure of Transportation Networks: Relationships between Network Geometry and Regional Characteristics , number Nr. 84 in University of Chicago. Department of Geography (University of Chicago., 1963)

    1963

  28. [36]

    G., A New Measure of Rank Correlation, Biometrika 30 (1938) 81–93, doi:10.2307/2332226

    Kendall, M. G., A New Measure of Rank Correlation, Biometrika 30 (1938) 81–93, doi:10.2307/2332226

    1938 doi

  29. [37]

    Kipf, T. N. and Welling, M., Variational Graph Auto-Encoders (2016), doi:10.48550/ September 3, 2025 17:43 manuscript 30 Chen Zhang and J¨ urgen Hackl arXiv.1611.07308

    2016 arXiv

  30. [38]

    Kirkley, A., Barbosa, H., Barthelemy, M., and Ghoshal, G., From the betweenness centrality in street networks to structural invariants in random planar graphs,Nature Communications 9 (2018) 2501, doi:10.1038/s41467-018-04978-z

    2018 doi

  31. [39]

    P., Moreno, Y., and Porter, M

    Kivel¨ a, M., Arenas, A., Barthelemy, M., Gleeson, J. P., Moreno, Y., and Porter, M. A., Multilayer networks, Journal of Complex Networks 2 (2014) 203–271, doi: 10.1093/comnet/cnu016

    2014 doi

  32. [40]

    and Leibler, R

    Kullback, S. and Leibler, R. A., On Information and Sufficiency, The Annals of Math- ematical Statistics 22 (1951) 79–86

    1951

  33. [41]

    Lambiotte, R., Rosvall, M., and Scholtes, I., From networks to optimal higher- order models of complex systems, Nature Physics 15 (2019) 313–320, doi:10.1038/ s41567-019-0459-y

    2019

  34. [42]

    J., An Algorithm for the Discrete Network Design Problem, Transporta- tion Science 9 (1975) 183–199, doi:10.1287/trsc.9.3.183

    Leblanc, L. J., An Algorithm for the Discrete Network Design Problem, Transporta- tion Science 9 (1975) 183–199, doi:10.1287/trsc.9.3.183

    1975 doi

  35. [43]

    C., Understanding individual routing behaviour, Journal of The Royal Society Interface 13 (2016) 20160021, doi:10.1098/rsif.2016.0021

    Lima, A., Stanojevic, R., Papagiannaki, D., Rodriguez, P., and Gonz´ alez, M. C., Understanding individual routing behaviour, Journal of The Royal Society Interface 13 (2016) 20160021, doi:10.1098/rsif.2016.0021

    2016

  36. [44]

    T., Prioritizing transportation network recovery using a resilience measure, Sustainable and Resilient Infrastructure 7 (2022) 70–81, doi:10.1080/23789689.2019.1708180

    Liu, Y., McNeil, S., Hackl, J., and Adey, B. T., Prioritizing transportation network recovery using a resilience measure, Sustainable and Resilient Infrastructure 7 (2022) 70–81, doi:10.1080/23789689.2019.1708180

    2022

  37. [45]

    Lov´ asz, L., Random Walks on Graphs: A Survay, Technical Report YALEU/DCS/TR-1029, Yeale University, New Haven, CT, US (1994)

    1994

  38. [46]

    Mahnke, R., Kaupuˇ zs, J., and Lubashevsky, I., Probabilistic description of traffic flow, Physics Reports 408 (2005) 1–130, doi:10.1016/j.physrep.2004.12.001

    2005 doi

  39. [47]

    J., Addison, J

    Manley, E. J., Addison, J. D., and Cheng, T., Shortest path or anchor-based route choice: A large-scale empirical analysis of minicab routing in London, Journal of Transport Geography 43 (2015) 123–139, doi:10.1016/j.jtrangeo.2015.01.006

    2015 doi

  40. [48]

    A., and Lambiotte, R., Random walks and diffusion on net- works, Physics Reports 716–717 (2017) 1–58, doi:10.1016/j.physrep.2017.07.007

    Masuda, N., Porter, M. A., and Lambiotte, R., Random walks and diffusion on net- works, Physics Reports 716–717 (2017) 1–58, doi:10.1016/j.physrep.2017.07.007

    2017 doi

  41. [49]

    T., Data-driven estimation of deterioration curves: A railway supporting structures case study, Infrastructure Asset Management 9 (2022) 3–17, doi:10.1680/jinam.21.00006

    Moghtadernejad, S., Huber, G., Hackl, J., and Adey, B. T., Data-driven estimation of deterioration curves: A railway supporting structures case study, Infrastructure Asset Management 9 (2022) 3–17, doi:10.1680/jinam.21.00006

    2022 doi

  42. [50]

    and Fl¨ otter¨ od, G., Agent-based traffic assignment: Going from trips to behavioral travelers (2009)

    Nagel, K. and Fl¨ otter¨ od, G., Agent-based traffic assignment: Going from trips to behavioral travelers (2009)

    2009

  43. [51]

    Newman, M. E. J., The Structure and Function of Complex Networks, SIAM Review 45 (2003) 167–256, doi:10.1137/S003614450342480

    2003 doi

  44. [52]

    Nguyen, J., Powers, S. T., Urquhart, N., Farrenkopf, T., and Guckert, M., An overview of agent-based traffic simulators, Transportation Research Interdisciplinary Perspec- tives 12 (2021) 100486, doi:10.1016/j.trip.2021.100486

    2021

  45. [53]

    NVIDIA, cuGraph, RAPIDS (2025)

    2025

  46. [54]

    Page, L., Brin, S., Motwani, R., and Winograd, T., The PageRank citation ranking: Bringing order to the Web, in Proceedings of the 7th International World Wide Web Conference (Brisbane, Australia, 1998), pp. 161–172

    1998

  47. [55]

    Pearson, K., The Problem of the Random Walk, Nature 72 (1905) 342–342, doi: 10.1038/072342a0

    1905 doi

  48. [56]

    M., and Brown, C

    Pell, J., Hintze, A., Canino-Koning, R., Howe, A., Tiedje, J. M., and Brown, C. T., Scaling metagenome sequence assembly with probabilistic de Bruijn graphs, Proceedings of the National Academy of Sciences 109 (2012) 13272–13277, doi: 10.1073/pnas.1121464109. September 3, 2025...

    2012 doi

  49. [57]

    J., and Schweitzer, F., Applications of higher-order models, in Higher-Order Graph Models: From Theo- retical Foundations to Machine Learning (Dagstuhl Seminar 21352), eds

    Perri, V., DiBona, G., Gallo, L., Gote, C., Hackl, J., Higham, D. J., and Schweitzer, F., Applications of higher-order models, in Higher-Order Graph Models: From Theo- retical Foundations to Machine Learning (Dagstuhl Seminar 21352), eds. Eliassi-Rad, T., Latora, V., Rosvall, ...

    2021

  50. [58]

    51:1–51:21

    Qarkaxhija, L., Perri, V., and Scholtes, I., De Bruijn Goes Neural: Causality-Aware Graph Neural Networks for Time Series Data on Dynamic Graphs, in Proceedings of the First Learning on Graphs Conference (PMLR, 2022), ISSN 2640-3498, pp. 51:1–51:21

    2022

  51. [59]

    and Marschall, T., MBG: Minimizer-based sparse de Bruijn Graph construction, Bioinformatics 37 (2021) 2476–2478, doi:10.1093/bioinformatics/ btab004

    Rautiainen, M. and Marschall, T., MBG: Minimizer-based sparse de Bruijn Graph construction, Bioinformatics 37 (2021) 2476–2478, doi:10.1093/bioinformatics/ btab004

    2021 doi

  52. [60]

    V., Lancichinetti, A., West, J

    Rosvall, M., Esquivel, A. V., Lancichinetti, A., West, J. D., and Lambiotte, R., Mem- ory in network flows and its effects on spreading dynamics and community detection, Nature Communications 5 (2014) 4630, doi:10.1038/ncomms5630

    2014 doi

  53. [61]

    T., and Lambiotte, R., Using higher-order Markov models to reveal flow-based communities in networks, Scientific Reports 6 (2016) 23194, doi: 10.1038/srep23194

    Salnikov, V., Schaub, M. T., and Lambiotte, R., Using higher-order Markov models to reveal flow-based communities in networks, Scientific Reports 6 (2016) 23194, doi: 10.1038/srep23194

    2016 doi

  54. [62]

    1037–1046, doi:10.1145/3097983.3098145

    Scholtes, I., When is a Network a Network? Multi-Order Graphical Model Selection in Pathways and Temporal Networks, inProceedings of the 23rd ACM SIGKDD Interna- tional Conference on Knowledge Discovery and Data Mining , KDD ’17 (Association for Computing Machinery, New York, ...

    2017

  55. [63]

    J., and Schweitzer, F., Causality-driven slow-down and speed-up of diffusion in non-Markovian temporal networks, Nature Communications 5 (2014) 5024, doi:10.1038/ncomms6024

    Scholtes, I., Wider, N., Pfitzner, R., Garas, A., Tessone, C. J., and Schweitzer, F., Causality-driven slow-down and speed-up of diffusion in non-Markovian temporal networks, Nature Communications 5 (2014) 5024, doi:10.1038/ncomms6024

    2014 doi

  56. [64]

    Singer, P., Helic, D., Taraghi, B., and Strohmaier, M., Detecting Memory and Struc- ture in Human Navigation Patterns Using Markov Chain Models of Varying Order, PLOS ONE 9 (2014) e102070, doi:10.1371/journal.pone.0102070

    2014 doi

  57. [65]

    Stabler, B., Transportation Networks for Research. (2025)

    2025

  58. [66]

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., and Polosukhin, I., Attention is all you need, in Proceedings of the 31st Inter- national Conference on Neural Information Processing Systems , NIPS’17 (Curran Associates Inc., Red Hook, ...

    2017

  59. [67]

    and Hess, D

    Wang, C. and Hess, D. B., Role of Urban Big Data in Travel Behavior Research, Transportation Research Record 2675 (2021) 222–233, doi:10.1177/ 0361198120975029

    2021

  60. [68]

    M., and Van Mieghem, P., Betweenness centrality in a weighted network, Physical Review E 77 (2008) 046105, doi:10.1103/PhysRevE.77

    Wang, H., Hernandez, J. M., and Van Mieghem, P., Betweenness centrality in a weighted network, Physical Review E 77 (2008) 046105, doi:10.1103/PhysRevE.77. 046105

    2008 doi

  61. [69]

    Wang, Y., Yu, X., Guo, J., Papamichail, I., Papageorgiou, M., Zhang, L., Hu, S., Li, Y., and Sun, J., Macroscopic traffic flow modelling of large-scale freeway networks with field data verification: State-of-the-art review, benchmarking framework, and case studies using METANE...

    2022

  62. [70]

    Wang, Z., Scaglione, A., and Thomas, R. J., Generating Statistically Correct Ran- dom Topologies for Testing Smart Grid Communication and Control Networks,IEEE Transactions on Smart Grid 1 (2010) 28–39, doi:10.1109/TSG.2010.2044814

    2010

  63. [71]

    Watts, D. J. and Strogatz, S. H., Collective dynamics of ‘small-world’ networks, Nature 393 (1998) 440–442, doi:10.1038/30918. September 3, 2025 17:43 manuscript 32 Chen Zhang and J¨ urgen Hackl

    1998 doi

  64. [72]

    E., Lacasa, L., Mill´ an, A

    Williams, O. E., Lacasa, L., Mill´ an, A. P., and Latora, V., The shape of mem- ory in temporal networks, Nature Communications 13 (2022) 499, doi:10.1038/ s41467-022-28123-z

    2022

  65. [73]

    Wu, X., Cao, W., Wang, J., Zhang, Y., Yang, W., and Liu, Y., A spatial interaction incorporated betweenness centrality measure, PLoS ONE 17 (2022) e0268203, doi: 10.1371/journal.pone.0268203

    2022 doi

  66. [74]

    Xiang, N., Wang, Q., and You, M., Estimation and update of betweenness centrality with progressive algorithm and shortest paths approximation, Scientific Reports 13 (2023) 17110, doi:10.1038/s41598-023-44392-0

    2023 doi

  67. [75]

    L., and Chawla, N

    Xu, J., Wickramarathne, T. L., and Chawla, N. V., Representing higher-order de- pendencies in networks, Science Advances 2 (2016) e1600028, doi:10.1126/sciadv. 1600028

    2016 doi

  68. [76]

    and Li, G., A Survey of Traffic Prediction: From Spatio-Temporal Data to Intelligent Transportation, Data Science and Engineering 6 (2021) 63–85, doi: 10.1007/s41019-020-00151-z

    Yuan, H. and Li, G., A Survey of Traffic Prediction: From Spatio-Temporal Data to Intelligent Transportation, Data Science and Engineering 6 (2021) 63–85, doi: 10.1007/s41019-020-00151-z

    2021 doi

  69. [77]

    and Levinson, D., Do People Use the Shortest Path? An Empirical Test of Wardrop’s First Principle, PLOS ONE 10 (2015) e0134322, doi:10.1371/journal

    Zhu, S. and Levinson, D., Do People Use the Shortest Path? An Empirical Test of Wardrop’s First Principle, PLOS ONE 10 (2015) e0134322, doi:10.1371/journal. pone.0134322. September 3, 2025 17:43 manuscript Higher-Order Network Framework for Capturing Memory-Driven Mobility Dyn...

    2015 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.