Pith. sign in

REVIEW 2 major objections 54 references

TrajGenAgent: A Hierarchical LLM Agent for Human Mobility Trajectory Generation

T0 review · 2 major / 0 minor · reviewed 2026-06-27 · grok-4.3

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

Pith's one-line read TrajGenAgent generates realistic human mobility trajectories with a hierarchical LLM agent without any model fine-tuning.

desk verdict The hierarchical split between LLM activity synthesis and deterministic grounding is a sensible practical move, but the abstract supplies no numbers or details so the claimed gains stay unverified. read the letter →

arxiv 2606.12657 v1 pith:ULU5XG62 submitted 2026-06-10 cs.AI cs.DBcs.RO

Siyu Li , Toan Tran , Lingyi Zhao , Khurram Shafique , Li Xiong This is my paper
classification cs.AIcs.DBcs.RO
keywords humanmobilitytrajectorygenerationLLMagentsyntheticdataanomalydetectionin-contextlearningspatiotemporalfidelity
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 presents TrajGenAgent as a framework that creates synthetic movement paths for people by first letting an LLM build a daily activity sequence drawn from past examples through in-context learning, then applying fixed rules to assign real places, distances, and times. This two-part structure aims to combine the LLM's reasoning flexibility with precise spatiotemporal details while skipping the expense of retraining the model. The authors also introduce an evaluation method that uses two anomaly detectors to judge whether the paths match expected behavior and meaning, not just overall statistics. Such synthetic data matters for fields like city planning and epidemic tracking where real location records are limited by cost and privacy rules. Experiments on standard and large simulated sets indicate gains in location accuracy, activity sense, and person-specific patterns compared with prior neural and LLM approaches.

What carries the argument

The two-stage orchestrator-worker design, in which the LLM performs activity-chain synthesis via in-context learning and a deterministic workflow performs grounding through POI retrieval, location selection, travel-time propagation, and duration estimation.

What would settle it

A side-by-side run of TrajGenAgent against real held-out trajectories where the anomaly detectors score the generated paths lower than the real ones on behavioral or semantic plausibility would falsify the improvement claim.

Watch

Extended reading notes

Core claim

TrajGenAgent is a semantic-aware hierarchical LLM-agent framework for human mobility trajectory generation without model fine-tuning. It uses a two-stage orchestrator-worker design: an LLM first synthesizes an individual- and weekday-conditioned activity chain from historical evidence via in-context learning, and a deterministic workflow then grounds each activity into a complete visit using personalized POI retrieval, distance-aware location selection, kinematics-aware travel-time propagation, and LLM-based duration estimation. To evaluate realism beyond aggregate spatiotemporal statistics, the work introduces an anomaly-detection-based evaluation framework using two complementary detectors

Load-bearing premise

The two complementary anomaly detectors reliably measure behavioral and semantic plausibility beyond aggregate spatiotemporal statistics, and the deterministic grounding rules produce trajectories whose realism is not an artifact of the chosen POI retrieval or travel-time propagation heuristics.

Editorial extensions

If this is right

  • Synthetic trajectories achieve higher spatiotemporal fidelity than neural or LLM baselines while using no parameter updates.
  • The generated paths show improved semantic coherence and individual-specific behavioral realism on both benchmark and large-scale simulation data.
  • The anomaly-detection evaluation framework provides a way to check plausibility beyond standard aggregate statistics.
  • The approach preserves zero-shot reasoning capacity by avoiding fine-tuning.

Reading between the lines

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

  • The method could lower barriers to obtaining usable mobility data for privacy-regulated applications such as epidemic modeling.
  • Replacing the deterministic grounding step with learned components might test whether the current gains depend on the fixed rules.
  • The orchestrator-worker split could be tested on other sequential generation problems that mix flexible planning with precise constraints.
  • Scaling the same workflow to multi-city or multi-year datasets would reveal whether the in-context learning step remains stable.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 0 minor

Summary. The paper proposes TrajGenAgent, a semantic-aware hierarchical LLM-agent framework for human mobility trajectory generation without model fine-tuning. It uses a two-stage orchestrator-worker design: an LLM synthesizes an individual- and weekday-conditioned activity chain from historical evidence via in-context learning, followed by a deterministic workflow that grounds each activity into a complete visit using personalized POI retrieval, distance-aware location selection, kinematics-aware travel-time propagation, and LLM-based duration estimation. To evaluate realism beyond aggregate spatiotemporal statistics, the authors introduce an anomaly-detection-based evaluation framework using two complementary detectors to assess behavioral and semantic plausibility. Experiments on benchmark and large-scale simulation datasets are claimed to demonstrate improvements in spatiotemporal fidelity, semantic coherence, and individual-specific behavioral realism over neural and LLM-based baselines.

Significance. If the central claims hold and the evaluation framework is shown to be non-circular, the work could offer a computationally efficient, zero-shot alternative to fine-tuned neural generators for synthetic mobility data, with applications in transportation, urban planning, and epidemic modeling. The combination of in-context learning for activity chains with deterministic grounding rules is a clear strength, as is the explicit avoidance of parameter updates. However, the significance is tempered by the need to verify that reported gains reflect genuine improvements in LLM-driven reasoning rather than artifacts of the grounding heuristics.

major comments (2)
  1. [Evaluation framework / anomaly detectors] The evaluation framework (described in the abstract and evaluation section) introduces two complementary anomaly detectors specifically to measure behavioral and semantic plausibility 'beyond aggregate spatiotemporal statistics,' yet the grounding workflow already embeds strong priors via POI retrieval, distance-aware selection, kinematics-aware propagation, and duration estimation. The manuscript provides no evidence that the detectors are trained or calibrated independently of these heuristics or of the input mobility patterns; without such separation, improved detector scores risk being circular with the deterministic rules rather than validating superior activity-chain realism.
  2. [Abstract / Experiments] The abstract states that 'experiments on benchmark and large-scale simulation datasets show that TrajGenAgent improves spatiotemporal fidelity, semantic coherence, and individual-specific behavioral realism,' but supplies no quantitative metrics, baseline details, dataset sizes, error bars, or statistical significance tests. This leaves the central empirical claim without visible supporting evidence, making it impossible to assess whether the reported gains are load-bearing or merely descriptive.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our work. We address each major comment point by point below, providing clarifications and indicating where revisions will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Evaluation framework / anomaly detectors] The evaluation framework (described in the abstract and evaluation section) introduces two complementary anomaly detectors specifically to measure behavioral and semantic plausibility 'beyond aggregate spatiotemporal statistics,' yet the grounding workflow already embeds strong priors via POI retrieval, distance-aware selection, kinematics-aware propagation, and duration estimation. The manuscript provides no evidence that the detectors are trained or calibrated independently of these heuristics or of the input mobility patterns; without such separation, improved detector scores risk being circular with the deterministic rules rather than validating superior activity-chain realism.

    Authors: The two anomaly detectors are trained solely on held-out real trajectories from the benchmark datasets via cross-validation, learning distributions of authentic behavioral sequences and semantic activity patterns without exposure to any generated trajectories or the grounding heuristics. The heuristics operate exclusively in the synthesis stage and do not influence detector parameters or features. We will add an explicit subsection in the revised evaluation section documenting the training data splits, feature definitions, and independence verification to address this concern directly. revision: yes

  2. Referee: [Abstract / Experiments] The abstract states that 'experiments on benchmark and large-scale simulation datasets show that TrajGenAgent improves spatiotemporal fidelity, semantic coherence, and individual-specific behavioral realism,' but supplies no quantitative metrics, baseline details, dataset sizes, error bars, or statistical significance tests. This leaves the central empirical claim without visible supporting evidence, making it impossible to assess whether the reported gains are load-bearing or merely descriptive.

    Authors: Abstracts conventionally provide high-level summaries; the detailed quantitative results—including specific metric improvements (e.g., anomaly detection rates, spatiotemporal fidelity scores), baseline comparisons, dataset sizes and sources, error bars, and statistical significance tests—are fully reported in the Experiments section. We will revise the abstract to incorporate key numerical highlights from those results for improved clarity. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity detected

full rationale

The paper describes a hierarchical LLM-agent method relying on in-context learning for activity chain synthesis followed by deterministic grounding workflows (POI retrieval, distance-aware selection, kinematics-aware propagation, duration estimation) and an anomaly-detection evaluation framework. No equations, fitted parameters, or self-citations are referenced in the provided text that would reduce any claimed result to its inputs by construction. The central claims rest on the described workflow and external benchmarks rather than self-definitional or fitted-input patterns. This is a self-contained method description without load-bearing reductions of the enumerated kinds.

Assumptions & free parameters 0 free parameters · 1 assumptions · 0 invented entities

The framework depends on the unverified capability of LLMs to produce realistic activity chains via in-context learning and on the adequacy of the chosen deterministic rules for producing plausible trajectories; no free parameters or invented entities are introduced in the abstract.

assumptions (1)
  • domain assumption LLM in-context learning from historical evidence produces activity chains that are sufficiently realistic for downstream grounding
    Invoked for the orchestrator stage that synthesizes individual- and weekday-conditioned chains.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TrajGenAgent: A Hierarchical LLM Agent for Human Mobility Trajectory Generation." pith.science (2026). https://pith.science/paper/ULU5XG62

@misc{pith2026260612657,
  author       = {Pith},
  title        = {Pith review of: TrajGenAgent: A Hierarchical LLM Agent for Human Mobility Trajectory Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ULU5XG62}},
  note         = {Machine review of arXiv:2606.12657}
}
read the original abstract

Human mobility data is important for transportation, urban planning, and epidemic control, but large-scale trajectory collection is often costly and privacy-constrained, motivating realistic synthetic trajectory generation. Existing LLM-based generators typically rely on either prompt engineering, which preserves zero-shot reasoning but lacks fine-grained spatiotemporal grounding, or trajectory-level fine-tuning, which improves statistical precision but incurs substantial computational cost and may weaken general reasoning. We propose TrajGenAgent, a semantic-aware hierarchical LLM-agent framework for human mobility trajectory generation without model fine-tuning. TrajGenAgent uses a two-stage orchestrator-worker design: an LLM first synthesizes an individual- and weekday-conditioned activity chain from historical evidence via in-context learning, and a deterministic workflow then grounds each activity into a complete visit using personalized POI retrieval, distance-aware location selection, kinematics-aware travel-time propagation, and LLM-based duration estimation. To evaluate realism beyond aggregate spatiotemporal statistics, we introduce an anomaly-detection-based evaluation framework using two complementary detectors to assess behavioral and semantic plausibility. Experiments on benchmark and large-scale simulation datasets show that TrajGenAgent improves spatiotemporal fidelity, semantic coherence, and individual-specific behavioral realism over representative neural and LLM-based baselines, while avoiding parameter updates.

Figures

Figures reproduced from arXiv: 2606.12657 by the authors.

Figure 1
Figure 1. The TrajGenAgent framework. A hierarchical orchestrator–worker LLM agent workflow, where a LLM orchestrator [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The illustration of Stage 1 LLM semantic planning for [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 7 canonical work pages

  1. [1]

    Human mobility data in the covid-19 pandemic: characteristics, applications, and challenges,

    T. Hu, S. Wang, B. She, M. Zhang, X. Huang, Y . Cui, J. Khuri, Y . Hu, X. Fu, X. Wanget al., “Human mobility data in the covid-19 pandemic: characteristics, applications, and challenges,”International Journal of Digital Earth, vol. 14, no. 9, pp. 1126–1147, 2021

    2021

  2. [2]

    A tale of one city: Using cellular network data for urban planning,

    R. A. Becker, R. Caceres, K. Hanson, J. M. Loh, S. Urbanek, A. Var- shavsky, and C. V olinsky, “A tale of one city: Using cellular network data for urban planning,”IEEE Pervasive Computing, vol. 10, no. 4, pp. 18–26, 2011

    2011

  3. [3]

    Human-centered trajectory tracking control for autonomous vehicles with driver cut-in behavior prediction,

    Y . Chen, C. Hu, and J. Wang, “Human-centered trajectory tracking control for autonomous vehicles with driver cut-in behavior prediction,” IEEE Transactions on Vehicular Technology, vol. 68, no. 9, pp. 8461– 8471, 2019

    2019

  4. [4]

    Mobility data science: Perspectives and challenges,

    M. e. a. Mokbel, “Mobility data science: Perspectives and challenges,” ACM Trans. Spatial Algorithms Syst., vol. 10, no. 2, jul 2024. [Online]. Available: https://doi.org/10.1145/3652158

    work page doi:10.1145/3652158 2024

  5. [5]

    Constructing a synthetic population of establishments for the simmobility microsimulation platform,

    D.-T. Le, G. Cernicchiaro, C. Zegras, and J. Ferreira, “Constructing a synthetic population of establishments for the simmobility microsimulation platform,”Transportation Research Procedia, vol. 19, pp. 81–93, 2016, transforming Urban Mobility. mobil.TUM 2016. International Scientific Conference on Mobility and Transport. Conference Proceedings. [Online]....

    2016

  6. [6]

    Learning to simulate human mobility,

    J. Feng, Z. Yang, F. Xu, H. Yu, M. Wang, and Y . Li, “Learning to simulate human mobility,” inProceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, 2020, pp. 3426–3433

    2020

  7. [7]

    Generating realistic and representative trajectories with mobility behavior clustering,

    H. Lin, S. Shaham, Y .-Y . Chiang, and C. Shahabi, “Generating realistic and representative trajectories with mobility behavior clustering,” in Proceedings of the 31st ACM International Conference on Advances in Geographic Information Systems, 2023, pp. 1–4

    2023

  8. [8]

    Cs- gan: Modality-aware trajectory generation via clustering-based sequence gan,

    M. Zhang, H. Lin, S. Takagi, Y . Cao, C. Shahabi, and L. Xiong, “Cs- gan: Modality-aware trajectory generation via clustering-based sequence gan,” in2023 24th IEEE International Conference on Mobile Data Management (MDM). IEEE, 2023, pp. 148–157

    2023

Show all 54 references
  1. [9]

    A non- parametric generative model for human trajectories

    K. Ouyang, R. Shokri, D. S. Rosenblum, and W. Yang, “A non- parametric generative model for human trajectories.” inIJCAI, vol. 18, 2018, pp. 3812–3817

    2018

  2. [10]

    Practical synthetic human trajectories generation based on variational point processes,

    Q. Long, H. Wang, T. Li, L. Huang, K. Wang, Q. Wu, G. Li, Y . Liang, L. Yu, and Y . Li, “Practical synthetic human trajectories generation based on variational point processes,” inProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023, pp. 4561–4571

    2023

  3. [11]

    A survey of large language models,

    W. X. Zhao, K. Zhang, J. Xie, J. Liu, Z. Li, Y . Shan, G. Yang, S. He, Z. Wang, Z. Liuet al., “A survey of large language models,”arXiv preprint arXiv:2303.18223, 2023

    2023 arXiv

  4. [12]

    Geo-llama: Leveraging llms for human mobility trajectory generation with constraints,

    S. Li, T. Tran, H. Lin, J. Krumm, C. Shahabi, L. Zhao, K. Shafique, and L. Xiong, “Geo-llama: Leveraging llms for human mobility trajectory generation with constraints,” in2025 26th IEEE International Confer- ence on Mobile Data Management (MDM). IEEE, 2025, pp. 20–31

    2025

  5. [13]

    Urban mobility as- sessment using llms,

    P. Bhandari, A. Anastasopoulos, and D. Pfoser, “Urban mobility as- sessment using llms,” inProceedings of the 32nd ACM International Conference on Advances in Geographic Information Systems, 2024, pp. 67–79

    2024

  6. [14]

    A study on individual spatiotemporal activity generation method using mcp-enhanced chain-of-thought large language models,

    Y . Zhang, Y . Hu, and D. Wang, “A study on individual spatiotemporal activity generation method using mcp-enhanced chain-of-thought large language models,”arXiv preprint arXiv:2506.10853, 2025

    2025

  7. [15]

    Large language models as urban residents: An llm agent framework for personal mobility generation,

    W. JIAWEI, R. Jiang, C. Yang, Z. Wu, R. Shibasaki, N. Koshizuka, C. Xiaoet al., “Large language models as urban residents: An llm agent framework for personal mobility generation,”Advances in Neural Information Processing Systems, vol. 37, pp. 124 547–124 574, 2024

    2024

  8. [16]

    Towards foun- dation model-based generation of human mobility trajectories,

    S. Li, T. Tran, L. Zhao, K. Shafique, and L. Xiong, “Towards foun- dation model-based generation of human mobility trajectories,” inThe inaugural ACM SIGSPATIAL International Workshop on Urban Mobility Foundation, 2025, p. 22

    2025

  9. [17]

    Lora: Low-rank adaptation of large language models,

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, L. Wang, and W. Chen, “Lora: Low-rank adaptation of large language models,” inInternational Conference on Learning Representations (ICLR), 2022

    2022

  10. [18]

    React: Synergizing reasoning and acting in language models,

    S. Yao, J. Yu, J. Zhao, K. Narasimhan, O. Etzioni, and Y . Choi, “React: Synergizing reasoning and acting in language models,”arXiv preprint arXiv:2210.03629, 2022

    2022 arXiv

  11. [19]

    Toolformer: Language models can teach themselves to use tools,

    T. Schick, J. Dwivedi-Yu, R. Dess `ı, R. Raileanu, M. Lomeli, L. Zettle- moyer, N. Cancedda, and T. Scialom, “Toolformer: Language models can teach themselves to use tools,” inAdvances in Neural Information Processing Systems, 2023

    2023

  12. [20]

    xLAM: A family of large action models to empower ai agent systems,

    J. Zhang, T. Lan, M. Zhu, Z. Liu, and C. Xiong, “xLAM: A family of large action models to empower ai agent systems,” 2024, https:// huggingface.co/Salesforce/Llama-xLAM-2-8b-fc-r

    2024

  13. [21]

    Agent ai with langgraph: A modular framework for enhancing machine translation using large language models,

    J. Wang and Z. Duan, “Agent ai with langgraph: A modular framework for enhancing machine translation using large language models,” 2024, arXiv preprint; LangGraph framework for modular agent orchestration

    2024

  14. [22]

    Icad: A self- supervised autoregressive approach for multi-context anomaly detection in human mobility data,

    B. Azarijoo, M. D. Siampou, J. Krumm, and C. Shahabi, “Icad: A self- supervised autoregressive approach for multi-context anomaly detection in human mobility data,” inProceedings of the 33rd ACM International Conference on Advances in Geographic Information Systems, 2025, pp. 595–606

    2025

  15. [23]

    Bestad: Behavior-aware spatio-temporal anomaly detection for human mobility data,

    J. Xie, J. Kim, Y .-Y . Chiang, L. Zhao, and K. Shafique, “Bestad: Behavior-aware spatio-temporal anomaly detection for human mobility data,” inProceedings of the 2nd ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection, 2025, pp. 56–59

    2025

  16. [24]

    Hermoupolis: a trajectory generator for simulating generalized mobility patterns,

    N. Pelekis, C. Ntrigkogias, P. Tampakis, S. Sideridis, and Y . Theodoridis, “Hermoupolis: a trajectory generator for simulating generalized mobility patterns,” inMachine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2013, Prague, Czech Republic,...

    2013

  17. [25]

    The timegeo modeling framework for urban mobility without travel surveys,

    S. Jiang, Y . Yang, S. Gupta, D. Veneziano, S. Athavale, and M. C. Gonz´alez, “The timegeo modeling framework for urban mobility without travel surveys,”Proceedings of the National Academy of Sciences, vol. 113, no. 37, pp. E5370–E5378, 2016

    2016

  18. [26]

    Predicting the next location: A recurrent model with spatial and temporal contexts,

    Q. Liu, S. Wu, L. Wang, and T. Tan, “Predicting the next location: A recurrent model with spatial and temporal contexts,” inProceedings of the AAAI conference on artificial intelligence, vol. 30, no. 1, 2016

    2016

  19. [27]

    Deep- move: Predicting human mobility with attentional recurrent networks,

    J. Feng, Y . Li, C. Zhang, F. Sun, F. Meng, A. Guo, and D. Jin, “Deep- move: Predicting human mobility with attentional recurrent networks,” inProceedings of the 2018 world wide web conference, 2018, pp. 1459– 1468

    2018

  20. [28]

    Generative adversarial networks,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial networks,” Communications of the ACM, vol. 63, no. 11, pp. 139–144, 2020

    2020

  21. [29]

    Seqgan: Sequence generative adversarial nets with policy gradient,

    L. Yu, W. Zhang, J. Wang, and Y . Yu, “Seqgan: Sequence generative adversarial nets with policy gradient,” inProceedings of the AAAI conference on artificial intelligence, vol. 31, no. 1, 2017

    2017

  22. [30]

    Neural point process for learning spatiotemporal event dynamics,

    Z. Zhou, X. Yang, R. Rossi, H. Zhao, and R. Yu, “Neural point process for learning spatiotemporal event dynamics,” inLearning for Dynamics and Control Conference. PMLR, 2022, pp. 777–789

    2022

  23. [31]

    Large language models as urban residents: An llm agent framework for personal mobility generation,

    J. Wang, R. Jiang, C. Yang, Z. Wu, M. Onizuka, R. Shibasaki, N. Koshizuka, and C. Xiao, “Large language models as urban residents: An llm agent framework for personal mobility generation,”arXiv preprint arXiv:2402.14744, 2024

    2024

  24. [32]

    Chain of thought prompting elicits reasoning in large language models,

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, E. Chi, Q. Le, and D. Zhou, “Chain of thought prompting elicits reasoning in large language models,” inAdvances in Neural Information Processing Systems, 2022. [Online]. Available: https://arxiv.org/abs/2201.11903

    2022 arXiv

  25. [33]

    Gpt-4-oss: Open-source 120b parameter llm,

    O. S. Community, “Gpt-4-oss: Open-source 120b parameter llm,” https: //huggingface.co/TheBloke/GPT4-OSS-120B, 2024, accessed August 2025

    2024

  26. [34]

    Geographically masking health data to preserve confidentiality,

    M. P. Armstrong, G. Rushton, and D. L. Zimmerman, “Geographically masking health data to preserve confidentiality,”Statistics in medicine, vol. 18, no. 5, pp. 497–525, 1999

    1999

  27. [35]

    Ensuring confidentiality of geocoded health data: Assessing geographic masking strategies for individual-level data,

    P. A. Zandbergen, “Ensuring confidentiality of geocoded health data: Assessing geographic masking strategies for individual-level data,”Ad- vances in medicine, vol. 2014, no. 1, p. 567049, 2014

    2014

  28. [36]

    Large scale gps trajectory generation using map based on two stage gan,

    X. Wang, X. Liu, Z. Lu, and H. Yang, “Large scale gps trajectory generation using map based on two stage gan,”Journal of Data Science, vol. 19, no. 1, pp. 126–141, 2021

    2021

  29. [37]

    Generating mobility trajectories with retained data utility,

    C. Cao and M. Li, “Generating mobility trajectories with retained data utility,” inProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & data Mining, 2021, pp. 2610–2620

    2021

  30. [38]

    Difftraj: Generating gps trajectory with diffusion probabilistic model,

    Y . Zhu, Y . Ye, S. Zhang, X. Zhao, and J. Yu, “Difftraj: Generating gps trajectory with diffusion probabilistic model,”Advances in Neural Information Processing Systems, vol. 36, pp. 65 168–65 188, 2023

    2023

  31. [39]

    Where you like to go next: Successive point-of-interest recommendation,

    C. Cheng, H. Yang, M. R. Lyu, and I. King, “Where you like to go next: Successive point-of-interest recommendation,” inTwenty-Third international joint conference on Artificial Intelligence, 2013

    2013

  32. [40]

    Hst-lstm: A hierarchical spatial-temporal long- short term memory network for location prediction

    D. Kong and F. Wu, “Hst-lstm: A hierarchical spatial-temporal long- short term memory network for location prediction.” inIjcai, vol. 18, no. 7, 2018, pp. 2341–2347

    2018

  33. [41]

    Where to go next: Modeling long-and short-term user preferences for point-of- interest recommendation,

    K. Sun, T. Qian, T. Chen, Y . Liang, Q. V . H. Nguyen, and H. Yin, “Where to go next: Modeling long-and short-term user preferences for point-of- interest recommendation,” inProceedings of the AAAI conference on artificial intelligence, vol. 34, no. 01, 2020, pp. 214–221

    2020

  34. [42]

    Stan: Spatio-temporal attention network for next location recommendation,

    Y . Luo, Q. Liu, and Z. Liu, “Stan: Spatio-temporal attention network for next location recommendation,” inProceedings of the web conference 2021, 2021, pp. 2177–2185

    2021

  35. [43]

    Next point- of-interest recommendation with inferring multi-step future preferences

    L. Zhang, Z. Sun, Z. Wu, J. Zhang, Y . S. Ong, and X. Qu, “Next point- of-interest recommendation with inferring multi-step future preferences.” inIJCAI, 2022, pp. 3751–3757

    2022

  36. [44]

    Stp-udgat: Spatial-temporal-preference user dimensional graph attention network for next poi recommendation,

    N. Lim, B. Hooi, S.-K. Ng, X. Wang, Y . L. Goh, R. Weng, and J. Varadarajan, “Stp-udgat: Spatial-temporal-preference user dimensional graph attention network for next poi recommendation,” inProceedings of the 29th ACM International conference on information & knowledge managem...

    2020

  37. [45]

    Getnext: trajectory flow map enhanced transformer for next poi recommendation,

    S. Yang, J. Liu, and K. Zhao, “Getnext: trajectory flow map enhanced transformer for next poi recommendation,” inProceedings of the 45th International ACM SIGIR Conference on research and development in information retrieval, 2022, pp. 1144–1153

    2022

  38. [46]

    Large language models for next point-of-interest recommendation,

    P. Li, M. de Rijke, H. Xue, S. Ao, Y . Song, and F. D. Salim, “Large language models for next point-of-interest recommendation,” inProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2024, pp. 1463– 1472

    2024

  39. [47]

    Qwen3 technical report,

    A. Yang, A. Li, B. Yang, B. Zhang, B. Hui, B. Zheng, B. Yu, C. Gao, C. Huang, C. Lvet al., “Qwen3 technical report,”arXiv preprint arXiv:2505.09388, 2025

    2025 arXiv

  40. [48]

    Efficient memory management for large language model serving with pagedattention,

    W. Kwon, Z. Li, S. Zhuang, Y . Sheng, L. Zheng, C. H. Yu, J. Gonzalez, H. Zhang, and I. Stoica, “Efficient memory management for large language model serving with pagedattention,” inProceedings of the 29th symposium on operating systems principles, 2023, pp. 611–626

    2023

  41. [49]

    Numosim: A synthetic mobility dataset with anomaly detection benchmarks,

    C. Stanford, S. Adari, X. Liao, Y . He, Q. Jiang, C. Kuai, J. Ma, E. Tung, Y . Qian, L. Zhaoet al., “Numosim: A synthetic mobility dataset with anomaly detection benchmarks,” inProceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection, 2024, ...

    2024

  42. [50]

    A scalable multi-modal framework for high-fidelity distributed human mobility simulations,

    S. B. Yoginath, N. Ahmad, C. Gunaratne, L. Amichi, J.-S. Kim, A. Burger, H. Xu, B. Bishnoi, S. C. Christopher, and G. M. Thakur, “A scalable multi-modal framework for high-fidelity distributed human mobility simulations,” inProceedings of the 8th ACM SIGSPATIAL International W...

    2025

  43. [51]

    Controllable visit trajectory generation with spatiotemporal constraints,

    H. Lin, J. Krumm, C. Shahabi, and L. Xiong, “Controllable visit trajectory generation with spatiotemporal constraints,” inProceedings of the 2024 IEEE International Conference on Data Mining (ICDM). IEEE, 2024

    2024

  44. [52]

    Gradient-based learning applied to document recognition,

    Y . LeCun, L. Bottou, Y . Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,”Proceedings of the IEEE, vol. 86, no. 11, pp. 2278–2324, 1998

    1998

  45. [53]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural computation, vol. 9, no. 8, pp. 1735–1780, 1997

    1997

  46. [54]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Advances in Neural Information Processing Systems, vol. 30, 2017

    2017

Pith tools

Reviewed June 27, 2026 · model on record in the stance chip above.