pith. sign in

arxiv: 2607.00907 · v1 · pith:NH7C3IQPnew · submitted 2026-07-01 · 📊 stat.AP

Beyond the Flow: A Bayesian Latent Clustering Framework for Shared Micro-mobility Users in Venice

Vanshika Keshwani , Stefano Mazzuco This is my paper

Pith reviewed 2026-07-02 03:04 UTC · model grok-4.3

classification 📊 stat.AP
keywords Bayesian mixture modelmicro-mobilityuser clusteringlatent profilestrip data analysiscategorical observationsVenice mobility
0
0 comments X

The pith

A Bayesian finite mixture model clusters shared micro-mobility users directly from repeated trip observations into eight latent profiles.

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

The paper introduces a Bayesian finite mixture model that groups users according to their full sequences of categorical trip records instead of first collapsing those records into user-level averages. This keeps the original trip-level structure intact and treats cluster membership as a latent variable that is inferred jointly with the cluster-specific behavior parameters. The model is applied to a year of shared bike and e-bike trips in Venice, producing eight profiles that separate localized, commuter-oriented, tourist-oriented, central, and inter-zonal patterns. The approach is presented as a general, scalable tool for any large collection of repeated categorical observations in transportation or behavioral data.

Core claim

The analysis identifies eight distinct latent mobility profiles corresponding to localized, commuter-oriented, tourist-oriented, central, and inter-zonal travel behaviors. The model represents each user with a product-multinomial likelihood with latent cluster membership and is illustrated on over 220,000 trips by more than 11,000 recurrent users in Venice.

What carries the argument

Bayesian finite mixture model with product-multinomial likelihood and latent cluster membership for multivariate categorical count data

Load-bearing premise

The product-multinomial likelihood with latent cluster membership adequately represents the generative process for each user's sequence of categorical trip observations without substantial unmodeled dependence or selection effects in the recurrent-user subset.

What would settle it

Re-estimating the model on a held-out portion of the Venice trips or on an independent dataset from another city and obtaining a materially different number or interpretation of profiles would indicate the eight-profile structure is not stable.

Figures

Figures reproduced from arXiv: 2607.00907 by Stefano Mazzuco, Vanshika Keshwani.

Figure 1
Figure 1. Figure 1: The mainland regions in the Municipality of Venice [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Directed Acyclic Graph (DAG) representation of the product [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Vehicle pass distribution by cluster The duration of the trip taken also reveals differences in the formed clusters [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Trip-duration distribution by cluster Seasonality in determining user demand plays an integral role in under￾standing usage patterns across the year. Cluster 3 is strongly dominated by summer-season usage ( [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Seasonal distribution by cluster 13 [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Spatial network structures for latent mobility clusters [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Variational Bayes model selection by ELBO [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Hyperparameter sensitivity analysis. 6 Discussion and Conclusion The study identifies eight distinct latent user clusters within the Venice shared micromobility system using a Bayesian mixture of multinomial product models estimated via variational approach. The resulting clusters show substantial het￾erogeneity between users in terms of vehicle pass, duration of travel, and season preference. Users travel… view at source ↗
Figure 9
Figure 9. Figure 9: Confusion matrix for simulated results −82500 −82000 −81500 −81000 5 6 7 8 9 10 11 12 Number of clusters (K) ELBO [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Model selection from simulated data 24 [PITH_FULL_IMAGE:figures/full_fig_p024_10.png] view at source ↗
read the original abstract

The study on shared micro-mobility is based on trip modeling and user data. User segmentation in shared micromobility systems is traditionally studied by aggregating trip-level observations into user-specific summary measures before applying clustering techniques. Such aggregation can obscure trip-level variability and lead to ecological fallacies if results are interpreted as applying to individual records. We propose a Bayesian finite mixture model for multivariate categorical count data that clusters users directly from repeated trip-level observations while preserving the full categorical structure of individual travel behavior. This approach focuses on identifying heterogeneous mobility users from high-dimensional categorical trip behavior while accounting for uncertainty in cluster assignments. Users are the fundamental unit of analysis for exploring latent cluster patterns. The model represents each user with a product-multinomial likelihood with latent cluster membership. The methodology is illustrated using a one-year trip record of shared bikes and e-bikes from the Municipality of Venice, Italy, comprising over 220,000 trips made by more than 11,000 recurrent users. The analysis identifies eight distinct latent mobility profiles corresponding to localized, commuter-oriented, tourist-oriented, central, and inter-zonal travel behaviors. The proposed framework provides a flexible and computationally scalable approach for clustering repeated categorical observations and is readily applicable to other large-scale behavioral and transportation datasets.

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

3 major / 0 minor

Summary. The manuscript proposes a Bayesian finite mixture model with product-multinomial likelihood to cluster users of shared micro-mobility systems directly from repeated categorical trip observations, avoiding aggregation biases. Applied to a Venice dataset of over 220,000 trips by more than 11,000 recurrent users, the analysis identifies eight distinct latent mobility profiles corresponding to localized, commuter-oriented, tourist-oriented, central, and inter-zonal behaviors.

Significance. If the model is correctly specified, the number of clusters is robustly selected, and the independence assumption holds, the framework provides a principled Bayesian approach to user segmentation in high-dimensional categorical mobility data while quantifying assignment uncertainty. This could improve upon traditional summary-statistic clustering in transportation studies and extend to other behavioral datasets.

major comments (3)
  1. [Abstract] Abstract (model description): The product-multinomial likelihood is stated to represent each user's trips conditional on latent cluster membership, but no derivation, prior specifications, or posterior sampling details are supplied. Without these, the mapping from the 220,000 trips to the eight claimed profiles cannot be verified or reproduced.
  2. [Abstract] Abstract (model description): The finite mixture recovers eight profiles, yet the product-multinomial treats trips as conditionally i.i.d. given cluster. Sequential mobility data typically exhibits return-trip structure, time-of-day autocorrelation, and zone-transition dependence not encoded in the categorical categories; no robustness check or alternative specification (e.g., Markov dependence) is described, which directly affects the distinctness of the recovered profiles.
  3. [Abstract] Abstract (results): The number of clusters is reported as eight with no mention of the model-selection procedure (e.g., marginal likelihood, WAIC, or posterior predictive checks). This choice is load-bearing for the central claim of exactly eight distinct profiles.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify aspects of model specification and selection. We respond point-by-point below, indicating planned revisions where appropriate. The full manuscript provides technical details beyond the abstract, but we agree the abstract can be strengthened for standalone readability.

read point-by-point responses

  1. Referee: [Abstract] Abstract (model description): The product-multinomial likelihood is stated to represent each user's trips conditional on latent cluster membership, but no derivation, prior specifications, or posterior sampling details are supplied. Without these, the mapping from the 220,000 trips to the eight claimed profiles cannot be verified or reproduced.

    Authors: The abstract is intentionally concise. The full manuscript derives the product-multinomial likelihood in Section 2.1, specifies independent Dirichlet priors on the cluster-specific multinomial parameters and a Dirichlet prior on the mixing proportions in Section 2.2, and details MCMC sampling via Hamiltonian Monte Carlo in Stan in Section 2.3, with convergence diagnostics reported. To improve accessibility, we will revise the abstract to note that full model specification, priors, and sampling details appear in the Methods section. revision: partial

  2. Referee: [Abstract] Abstract (model description): The finite mixture recovers eight profiles, yet the product-multinomial treats trips as conditionally i.i.d. given cluster. Sequential mobility data typically exhibits return-trip structure, time-of-day autocorrelation, and zone-transition dependence not encoded in the categorical categories; no robustness check or alternative specification (e.g., Markov dependence) is described, which directly affects the distinctness of the recovered profiles.

    Authors: The product-multinomial formulation is chosen deliberately to model the marginal distribution of trip categories per user, enabling direct clustering on the full categorical structure without aggregation. This exchangeability assumption within users identifies stable behavioral profiles rather than sequential dynamics. While return trips and temporal autocorrelation are present in the data, extending to a Markov or autoregressive specification would require a substantially different likelihood and is beyond the scope of the current framework focused on user-level heterogeneity. The eight profiles align with domain-expected patterns (localized, commuter, tourist). We will add a paragraph in the Discussion acknowledging this assumption and its implications for profile interpretation. revision: partial

  3. Referee: [Abstract] Abstract (results): The number of clusters is reported as eight with no mention of the model-selection procedure (e.g., marginal likelihood, WAIC, or posterior predictive checks). This choice is load-bearing for the central claim of exactly eight distinct profiles.

    Authors: Cluster number K=8 was selected by minimizing WAIC across K=2 to K=15, with the minimum occurring at K=8; this is reported in Section 3.2 along with sensitivity checks using posterior predictive checks on trip-type distributions. We agree the abstract should reference the selection criterion. We will revise the abstract to state that the number of clusters was determined via WAIC. revision: yes

Circularity Check

0 steps flagged

No circularity: standard mixture model fitted to data yields empirical clusters

full rationale

The paper proposes a Bayesian finite mixture model with product-multinomial likelihood for user-level trip counts and applies it to the Venice dataset to recover eight latent profiles. Cluster membership and profile descriptions are obtained by posterior inference on the observed counts; no equation or step equates a claimed result to a fitted parameter by construction, renames an input, or relies on a self-citation chain for its validity. The derivation is the standard latent-class model applied to external data and is therefore self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on a standard Bayesian mixture modeling assumption plus a data-specific modeling choice whose justification is not supplied in the abstract.

free parameters (1)
  • number of clusters
    The value 8 is reported as the number of profiles recovered; selection criterion is not stated in the abstract.
axioms (1)
  • domain assumption Each user's trips are independent draws from a product-multinomial distribution conditional on latent cluster membership.
    This is the explicit likelihood stated for the finite mixture model.
invented entities (1)
  • latent mobility profiles no independent evidence
    purpose: To represent heterogeneous user travel behaviors
    Eight profiles are postulated as the latent structure; no external falsifiable evidence for their existence is mentioned.

pith-pipeline@v0.9.1-grok · 5751 in / 1231 out tokens · 47445 ms · 2026-07-02T03:04:14.967615+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

33 extracted references · 30 canonical work pages

  1. [1]

    Clustering Users and Non-Users of Shared and Private e-Bikes and e-Scooters across Three

    Adorean, Emanuel-Cristian and Nofre, Jordi and Moura, Filipe , date =. Clustering Users and Non-Users of Shared and Private e-Bikes and e-Scooters across Three. doi:10.1016/j.tranpol.2025.103875 , url =

    work page doi:10.1016/j.tranpol.2025.103875 2025

  2. [2]

    Anable, Jillian , date =. ‘. doi:10.1016/j.tranpol.2004.11.004 , url =

    work page doi:10.1016/j.tranpol.2004.11.004 2004

  3. [3]

    How Does Weather Affect Bikeshare Use?

    Bean, Richard and Pojani, Dorina and Corcoran, Jonathan , date =. How Does Weather Affect Bikeshare Use?. doi:10.1016/j.jtrangeo.2021.103155 , url =

    work page doi:10.1016/j.jtrangeo.2021.103155 2021

  4. [4]

    Beza, Abebe Dress and Demissie, Merkebe Getachew and Kattan, Lina , date =. A. doi:10.1177/03611981251350652 , url =

    work page doi:10.1177/03611981251350652

  5. [5]

    and Ng, Andrew Y

    Blei, David M. and Ng, Andrew Y. and Jordan, Michael I. , date =. Latent\_. doi:10.1162/jmlr.2003.3.4-5.993 , url =

    work page doi:10.1162/jmlr.2003.3.4-5.993 2003

  6. [6]

    Journal of the American Statistical Association 113(523), 1094–1111 (2018) https://doi.org/10.1080/01621459.2017.1307116

    Blei, David M. and Kucukelbir, Alp and McAuliffe, Jon D. , date =. Variational. doi:10.1080/01621459.2017.1285773 , url =

    work page doi:10.1080/01621459.2017.1285773 2017

  7. [7]

    Neighborhood Characteristics Encouraging Micromobility:

    Christoforou, Zoi and Psarrou Kalakoni, Anna Mariam and Farhi, Nadir , date =. Neighborhood Characteristics Encouraging Micromobility:. doi:10.1016/j.tbs.2023.02.002 , url =

    work page doi:10.1016/j.tbs.2023.02.002 2023

  8. [8]

    Fan, Aihua and Chen, Xumei and Wan, Tao , date =. How. doi:10.1155/2019/5426080 , url =

    work page doi:10.1155/2019/5426080 2019

  9. [9]

    Build It and Give ‘em Bikes, and They Will Come:

    Félix, Rosa and Cambra, Paulo and Moura, Filipe , date =. Build It and Give ‘em Bikes, and They Will Come:. doi:10.1016/j.cstp.2020.03.002 , url =

    work page doi:10.1016/j.cstp.2020.03.002 2020

  10. [10]

    Extracting Urban Patterns from Location-Based Social Networks , booktitle =

    Ferrari, Laura and Rosi, Alberto and Mamei, Marco and Zambonelli, Franco , date =. Extracting Urban Patterns from Location-Based Social Networks , booktitle =. doi:10.1145/2063212.2063226 , url =

    work page doi:10.1145/2063212.2063226

  11. [11]

    Fishman, Elliot and Washington, Simon and Haworth, Narelle , date =. Bike. doi:10.1080/01441647.2013.775612 , url =

    work page doi:10.1080/01441647.2013.775612 2013

  12. [12]

    Shared Micromobility in Multimodal Travel:

    Fu, Xingxing and Van Lierop, Dea and Ettema, Dick , date =. Shared Micromobility in Multimodal Travel:. doi:10.1016/j.cities.2024.105664 , url =

    work page doi:10.1016/j.cities.2024.105664 2024

  13. [13]

    Drivers and Barriers to Integrating Shared Micromobility with Public Transport

    Geržinič, Nejc and Van Hagen, Mark and Al-Tamimi, Hussein and Van Oort, Niels and Duives, Dorine , date =. Drivers and Barriers to Integrating Shared Micromobility with Public Transport. doi:10.1016/j.jcmr.2025.100090 , url =

    work page doi:10.1016/j.jcmr.2025.100090 2025

  14. [14]

    doi:10.1016/j.tranpol.2026.104042 , url =

    Assessment of Travel Behavior Dynamics among Young Demographic for Different Trip Purposes of Shared Micro-Mobility Services , author =. doi:10.1016/j.tranpol.2026.104042 , url =

    work page doi:10.1016/j.tranpol.2026.104042 2026

  15. [15]

    Analyzing

    Javaheri, Atusa and Pamidimukkala, Apurva and Kermanshachi, Sharareh and Rosenberger, Jay Michael and Hladik, Greg , date =. Analyzing. doi:10.1016/j.trpro.2025.10.091 , url =

    work page doi:10.1016/j.trpro.2025.10.091 2025

  16. [16]

    and Ghahramani, Zoubin and Jaakkola, Tommi S

    Jordan, Michael I. and Ghahramani, Zoubin and Jaakkola, Tommi S. and Saul, Lawrence K. , editor =. An. Learning in. doi:10.1007/978-94-011-5014-9_5 , url =

    work page doi:10.1007/978-94-011-5014-9_5

  17. [17]

    Gibbs Sampler and Coordinate Ascent Variational Inference:

    Lee, Se Yoon , date =. Gibbs Sampler and Coordinate Ascent Variational Inference:. doi:10.1080/03610926.2021.1921214 , url =

    work page doi:10.1080/03610926.2021.1921214 2021

  18. [18]

    Li, Weibo and Kamargianni, Maria , date =. A

  19. [19]

    doi:10.1016/j.tranpol.2025.04.025 , url =

    Investigating User Preferences for Dockless Bike- and Electric Bike-Sharing through Tracking Usage Patterns , author =. doi:10.1016/j.tranpol.2025.04.025 , url =

    work page doi:10.1016/j.tranpol.2025.04.025 2025

  20. [20]

    Shared Micro-Mobility Meets Bus:

    Liu, Xize and Chen, Jingxu and Chen, Xuewu and Ning, Jiang , date =. Shared Micro-Mobility Meets Bus:. doi:10.1016/j.tranpol.2025.103853 , url =

    work page doi:10.1016/j.tranpol.2025.103853 2025

  21. [21]

    Spatiotemporal Comparative Analysis of Scooter-Share and Bike-Share Usage Patterns in

    McKenzie, Grant , date =. Spatiotemporal Comparative Analysis of Scooter-Share and Bike-Share Usage Patterns in. doi:10.1016/j.jtrangeo.2019.05.007 , url =

    work page doi:10.1016/j.jtrangeo.2019.05.007 2019

  22. [22]

    McLachlan, Geoffrey and Peel, David , date =. Finite. doi:10.1002/0471721182 , url =

    work page doi:10.1002/0471721182

  23. [23]

    and Handy, Susan L

    Mohiuddin, Hossain and Fitch-Polse, Dillon T. and Handy, Susan L. , date =. Examining Market Segmentation to Increase Bike-Share Use and Enhance Equity:. doi:10.1016/j.tranpol.2023.10.021 , url =

    work page doi:10.1016/j.tranpol.2023.10.021 2023

  24. [24]

    , date =

    Murphy, Kevin P. , date =. Machine Learning: A Probabilistic Perspective , shorttitle =

  25. [25]

    Pobudzei, Maryna and Hoffmann, Silja , date =. User. 2024. doi:10.1109/IV55156.2024.10588699 , url =

    work page doi:10.1109/iv55156.2024.10588699 2024

  26. [26]

    Understanding Multimodal Mobility Patterns of Micromobility Users in Urban Environments: Insights from

    Roig-Costa, Oriol and Marquet, Oriol and Arranz-López, Aldo and Miralles-Guasch, Carme and Van Acker, Veronique , date =. Understanding Multimodal Mobility Patterns of Micromobility Users in Urban Environments: Insights from. doi:10.1007/s11116-024-10531-3 , url =

    work page doi:10.1007/s11116-024-10531-3

  27. [27]

    doi:10.1016/j.tranpol.2023.03.006 , url =

    What Are the Factors Affecting the Adoption and Use of Electric Scooter Sharing Systems from the End User's Perspective? , author =. doi:10.1016/j.tranpol.2023.03.006 , url =

    work page doi:10.1016/j.tranpol.2023.03.006 2023

  28. [28]

    Examining Spatio-Temporal Mobility Patterns of Bike-Sharing Systems: The Case of

    Talavera-Garcia, Ruben and Romanillos, Gustavo and Arias-Molinares, Daniela , date =. Examining Spatio-Temporal Mobility Patterns of Bike-Sharing Systems: The Case of. doi:10.1080/17445647.2020.1866697 , url =

    work page doi:10.1080/17445647.2020.1866697 2020

  29. [29]

    doi:10.1371/journal.pone.0238143 , url =

    Estimation of the Shared Mobility Demand Based on the Daily Regularity of the Urban Mobility and the Similarity of Individual Trips , author =. doi:10.1371/journal.pone.0238143 , url =

    work page doi:10.1371/journal.pone.0238143

  30. [30]

    and Jordan, Michael I

    Wainwright, Martin J. and Jordan, Michael I. , date =. Graphical. doi:10.1561/2200000001 , url =

    work page doi:10.1561/2200000001

  31. [31]

    Who Are the ‘Super-Users’ of Public Bike Share?

    Winters, Meghan and Hosford, Kate and Javaheri, Sana , date =. Who Are the ‘Super-Users’ of Public Bike Share?. doi:10.1016/j.pmedr.2019.100946 , url =

    work page doi:10.1016/j.pmedr.2019.100946 2019

  32. [32]

    Padgham, Mark , date =. Dodgr:. doi:10.32866/6945 , url =

    work page doi:10.32866/6945

  33. [33]

    Software: Practice and Experience , volume=

    Graph drawing by force-directed placement , author=. Software: Practice and Experience , volume=. 1991 , publisher=

    1991