Pith. sign in

REVIEW 4 major objections 4 minor 50 references

Traj-MLLM: Can Multimodal Large Language Models Reform Trajectory Data Mining?

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A frozen multimodal LLM, fed map-cropped images and structured text, can answer four trajectory-mining tasks with no training, the paper argues.

desk verdict Clever representation, but TTE and MP results likely reflect label leakage (durations in text, destination in map), not generalization. read the letter →

arxiv 2509.00053 v1 pith:L3ETR4QV submitted 2025-08-25 cs.MM cs.AIcs.CL

Shuo Liu , Di Yao , Yan Lin , Gao Cong , Jingping Bi This is my paper · ORCID
classification cs.MMcs.AIcs.CL
keywords trajectorydataminingmultimodallargelanguagemodelstraining-freeinferencemap-anchoredtokenizationtraveltimeestimationmobilitypredictionanomalydetectiontransportationmodeidentification
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 claims that a single frozen multimodal large language model can perform four standard trajectory mining tasks—travel time estimation, mobility prediction, anomaly detection, and transportation mode identification—across cities with no training or fine-tuning. To make this work, the authors convert each GPS trajectory into an interleaved sequence of map images and structured text descriptions, then hand a task prompt to the MLLM. They report large gains over task-specific and foundation models, for example cutting travel-time RMSE by over 48% on one city and more than doubling PR-AUC on a low-magnitude detour anomaly setting. A prompt-optimization loop using a few labeled seed trajectories produces data-invariant prompts that transfer across regions. The central question—whether MLLMs can reform trajectory mining—is answered affirmatively on these four tasks and four cities.

What carries the argument

The central mechanism is map-anchored tokenization combined with multiview trajectory modeling. Segmentation minimizes a cost function built from speed consistency, road-type changes, and segment length regularization; each resulting sub-trajectory becomes both a map-cropped image (with context elements filtered by a distance threshold) and a text block of summary statistics. Spatial views at global and local scales plus contextual views (POI, road network) are arranged in an ordered interleaved image-text sequence, letting the MLLM's sequential reasoning capture temporal dependencies without a trainable temporal encoder. Task adaptation rides on a prompt template of role, task description, domain knowledge, and output format, which is automatically refined over a handful of seed trajectories until it performs satisfactorily, then reused across cities.

What would settle it

Re-run travel time estimation and mobility prediction with the label-bearing pieces removed: drop the end timestamp and duration from every text block and crop the destination region out of the global map image, then compare the reported metrics. If accuracy collapses toward baseline chance, the original numbers came from reading the label in the input. A complementary check is to feed the text statistics alone, without any image, to the same MLLM and see how much of the reported travel-time and mobility improvement survives.

Watch

Extended reading notes

Core claim

Traj-MLLM claims to be the first general, training-free framework that uses multimodal LLMs for trajectory data mining. Raw GPS trajectories are segmented into semantically coherent sub-trajectories via a map-anchored dynamic program, rendered as map images with layered POI and road network context, and paired with structured statistics such as start and end times, duration, distance, and speeds. The interleaved image-text sequence preserves spatial, temporal, and contextual information, while a few-shot prompt optimization yields task prompts that transfer across regions. The claimed outcome is that the frozen MLLM outperforms state-of-the-art baselines by 48.05% on travel time estimation, 15.52% on mobility prediction, 51.52% on anomaly detection, and 1.83% on transportation mode identification, with ablation studies showing that the visual modality, semantic segmentation, interleaved order, POI context, and road network context all contribute to these results.

Load-bearing premise

The inputs given to the language model do not already contain the answer to the task: in particular, the global map image shows the full trajectory including the destination, and the text blocks state start and end timestamps and duration, so the reported travel-time and mobility numbers could be read off or arithmetically derived from the input rather than inferred from spatial semantics.

Editorial extensions

If this is right

  • A single frozen MLLM could replace task-specific and region-specific trained models for four trajectory tasks, requiring no training data for new cities.
  • Because the trajectory-to-sequence mapping is task-independent, new trajectory tasks can be added by writing a new task prompt and reusing the same multimodal representation.
  • Cross-city transfer is claimed to hold without retraining: the same optimized prompts work on Xi'an, Chengdu, Porto, and the GeoLife Beijing trajectories.
  • The released multimodal response dataset could serve as training data for building future MLLMs specialized for trajectory analysis.
  • The framework is extensible to additional contextual views, such as traffic lights, as long as they can be rendered as map layers and filtered by relevance.

Reading between the lines

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

  • If the reported numbers hold with properly masked inputs, the real contribution would be a test-time-compute substitute for model training: any trajectory task becomes a prompt, and performance would track the underlying MLLM's reasoning power, which the paper's own backbone comparison already hints at.
  • The prompt-optimization loop is de facto few-shot learning: the seed trajectories and their labels leak task information into the prompt, so the 'training-free' claim reduces to 'no gradient updates,' which matters for deployment but is not the same as zero-shot learning.
  • A natural testable extension is to probe how the distance threshold should scale with urban density; the paper's sensitivity analysis shows that too small a threshold can hide connecting roads and too large a threshold adds clutter, suggesting a dataset-dependent optimum that may not transfer to cities with sparse POI coverage.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes Traj-MLLM, a training-free framework that converts raw GPS trajectories into interleaved image-text sequences through map-anchored semantic segmentation and multiview modeling, and then uses a frozen multimodal large language model with task prompts optimized on a few seed trajectories to perform travel time estimation, anomaly detection, mobility prediction, and transportation mode identification. Experiments on Xian, Chengdu, Porto, and Geolife report large improvements over state-of-the-art baselines, and the authors release their code and a dataset of MLLM responses.

Significance. If the reported results were valid, Traj-MLLM would be a significant demonstration that a single frozen MLLM can generalize across cities and trajectory-mining tasks without parameter updates. The paper contributes a modular tokenization pipeline and publicly releases both code and a dataset of MLLM responses, which are valuable assets for the community. However, the current evaluation does not establish the central claim: the TTE and MP targets appear to be present in the multimodal inputs unless masking is applied, and no masking is described anywhere in the manuscript. In addition, the headline improvements are selected from the most favorable per-metric, per-dataset cells. The significance of the paper therefore depends on a repaired evaluation protocol.

major comments (4)
  1. [Section 4.1.2 and Appendix A.2] The TTE target is directly available in the text tokens. Equation (3) and Appendix A.2 specify that each sub-trajectory text token contains Start Time, End Time, and Duration in seconds. Since the TTE task is defined as predicting total travel time (Appendix A.3.2), the MLLM can obtain the answer by subtracting the first start time from the last end time or by summing the segment durations. The manuscript never states that timestamps or durations are masked for TTE, so the reported 48.05% improvement may reflect label extraction or arithmetic on the input rather than trajectory reasoning. The authors must either mask these fields and re-run the experiments, or demonstrate that performance is unaffected when they are removed.
  2. [Sections 4.1.2, 4.2.1, and Appendix A.3.2] The mobility prediction target is visible in the input. Mobility prediction is defined as predicting the destination region, but the global spatial view renders the entire trajectory and the interleaved text sequence includes the final sub-trajectory, whose endpoint is the destination. Section 4.1.2 additionally states that start and end points of each segment are marked with unique icons on the rendered map. Unless the destination is cropped from the images or the final segment is withheld from the text, the MLLM can read the answer directly from the global view or the last text token. No such masking is described, so the claimed 15.52% ACC@1 improvement does not currently constitute evidence of destination prediction.
  3. [Abstract and Tables 1-3] The headline percentages are cherry-picked from individual cells. The abstract's 48.05% for TTE is the RMSE improvement on Xian in Table 1; 15.52% for MP is the ACC@1 improvement on Porto; 51.52% for AD is the PR-AUC improvement for switch anomalies with μ=0.3 on Xian in Table 2; and 1.83% for TMI is the accuracy improvement on Geolife in Table 3. Many cells show far smaller gains, e.g., AD high-detour improvements of 0.20%, 0.72%, and 0.93%. The 'Improvement' rows are also computed against the best baseline per metric rather than a single consistent SOTA method. The authors should report all metrics across all datasets, state a primary metric per task in advance, and provide averages or ranges instead of selecting the most favorable numbers for the abstract.
  4. [Section 4.3.2] The claim that Traj-MLLM works 'without requiring any training data' is contradicted by the prompt optimization procedure. This section describes selecting labeled seed trajectories, using the true label ŷ to generate feedback from disagreement samples, and iteratively refining the prompt until it achieves satisfactory performance on the seed set. This is a supervised adaptation step that uses labeled data, even if it does not update model parameters. The authors must state how seed trajectories are selected, whether they are disjoint from the evaluation set, and should rephrase the claim to something like 'no model parameter training or fine-tuning' if that is what is intended.
minor comments (4)
  1. [Appendix A.3.2] The tasks section lists Accuracy, Precision, and PR-AUC for anomaly detection, but Table 2 reports only PR-AUC; please clarify which metrics are used and where the other metrics are reported.
  2. [Section 4.2] The contextual views are defined as C = {C_1, ..., C_K} in the first paragraph and then as C = {C_1, ..., C_Z} in Algorithm 1; please use a single index bound consistently.
  3. [Tables 1-3] No confidence intervals, standard deviations, or repeated runs are reported for the MLLM outputs. Given the stochasticity of MLLM decoding, the small 1.83% TMI improvement may be within run-to-run noise; please report variance across multiple inference runs or state the decoding temperature and sampling protocol.
  4. [Section 5.1] The paper states that 'over 80,000 trajectories' are selected, but does not specify how many trajectories are used for each task and dataset; please provide a per-task, per-dataset breakdown.

Circularity Check

4 steps flagged · score 8.0 of 10

TTE and MP labels are embedded in the multimodal input by construction; prompt optimization and per-city thresholds use labeled data, so the headline 'training-free' gains are partly answer extraction and fitted tuning.

  1. self definitional [Section 4.1.2 / Appendix A.2 / Appendix A.3.2 (TTE tokenization and task definition)]
    "Text Tokenization. For each sub-trajectory S_n, multiple statistical features are extracted from raw trajectory to generate a structured text description D_n. ... The components here include Time(·) for the start and end timestamps, Dist(·) for the total distance traveled ... Duration is the total time elapsed, calculated as (t_k−t_1) in seconds. ... Travel Time Estimation (TTE): A regression task to predict the total travel time of a given trajectory."

    By the paper's own tokenizer, every text token contains the TTE label: TTE asks for total travel time, and the global-view text token, which also passes through this tokenizer per Section 4.2.1, includes Start Time, End Time, and Duration for the whole trajectory. The paper nowhere describes masking or withholding the duration/timestamps. The MLLM can therefore output the duration field directly, or trivially sum segment durations, without generalizing from trajectory patterns. The reported 48.05% MAE improvement is thus consistent with answer extraction/arithmetic on the input rather than with learned trajectory-mining ability.

  2. self definitional [Section 4.1.2 Visual Tokenization / Section 4.2.1 / Appendix A.3.2 (MP)]
    "The start and end points of the segment are marked with unique icons to indicate direction. ... g_global takes T as a whole to generate the global view V_global. ... Mobility Prediction (MP): A classification task that predicts the destination region of a trajectory."

    The global view renders the entire trajectory, so the final endpoint—the destination—is visible in the image; local segment views also mark every segment's start and end. MP's label is the destination region. No cropping, masking, or removal of the final point is described. The model can therefore classify by reading the rendered endpoint. The reported 15.52% ACC@1 improvement is at least partly an input-reading effect, not evidence of a general mobility-prediction ability.

2 more flagged steps
  1. fitted input called prediction [Section 4.3 / Abstract (training-free claim)]
    "This module aims to address this challenge with a few labeled seed trajectories (e.g., less than 10) and conducts multi-rounds interaction with MLLMs to optimize the task prompts. ... Traj-MLLM achieves these superior performances without requiring any training data or fine-tuning the MLLM backbones."

    The prompt-optimization loop feeds true labels back into the MLLM and refines the prompt until it shows 'satisfactory performance on Seed.' This is supervised prompt fitting on labeled trajectories—a form of training on task data. The abstract's 'without requiring any training data' is contradicted by the paper's own protocol. The optimized prompt is a learned component fitted to task labels, so the pipeline is not training-free as claimed.

  2. fitted input called prediction [Section 5.7 / Appendix A.4 (theta sensitivity)]
    "Based on this analysis, we set θ tailored to the scale of each city's road network. For the Xian, Chengdu, and Geolife datasets, which feature similarly larger road networks, we selected θdistance = 100 meters. In contrast, for the Porto dataset with its smaller transportation network, we chose a more constrained value of θdistance = 50 meters."

    The sensitivity analysis (Figure 6) selects the threshold that yields the best MAE on each city, and those thresholds are then fixed as the method's settings for reporting results. This is per-dataset tuning on the evaluation task; the reported numbers are not parameter-free zero-shot results. While this alone does not force the predictions, it further undercuts the 'no training data / no tuning' claim by fitting a task-relevant hyperparameter to the evaluation data.

full rationale

The central claim—that a frozen MLLM outperforms SOTA across cities with no training—is undercut by the paper's own tokenizer. For TTE, the global-view text token includes Duration for the whole trajectory, which is exactly the regression target; for MP, the global rendering shows the final endpoint, which is the destination. No masking is described anywhere, so the headline TTE (48.05%) and MP (15.52%) improvements can be explained by reading the answer out of the input rather than by general trajectory reasoning. Additionally, prompt optimization uses labeled seed trajectories in a supervised feedback loop, contradicting 'without requiring any training data,' and the per-city θ is chosen after inspecting performance. AD and TMI results are less obviously answer-in-input, so the paper is not entirely vacuous, but the strongest empirical claims reduce to input inspection and fitted prompt/threshold choices. This is internal-validity circularity, not a disagreement with community consensus.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The framework relies on several tuned quantities: the context threshold, the seed trajectories for prompt optimization, the segmentation cost components, the view counts, and the map padding factor. The most consequential assumption is that feeding the full trajectory, with timestamps, duration, and destination visible, does not leak the task answer. The paper's own Appendix A.2 exposes the duration in the text template, so the TTE task may reduce to reading a field. No new physical entities are introduced.

free parameters (5)
  • theta context filtering threshold = 50m (Porto), 100m (Xian, Chengdu, Geolife)
    Set per dataset after sensitivity analysis; controls how many POIs and roads are rendered, directly affecting the visual input and task performance.
  • Seed set for prompt optimization = fewer than 10 labeled trajectories per task, exact number not specified
    Labels from these seeds are used to refine task prompts, so the final prompt is fitted to the task and dataset.
  • Segment cost components = three components: speed consistency, road-type changes, inverse length; no weights reported
    The semantic segmentation cost in Equation 1 sums three factors without stated weights or normalization, so segment boundaries depend on unspecified scales.
  • Number of spatial views K and contextual views Z = K=2 (global, local), Z=2 (POI, road)
    Design choice; the paper does not report a selection criterion for view counts, which could affect which information is captured.
  • Map clipping padding factor Delta = not specified
    The padding factor for the bounding box in visual tokenization is undefined; this changes the map context visible to the MLLM.
assumptions (4)
  • standard math The Haversine distance formula and the DP segmentation produce semantically meaningful sub-trajectories.
    Used in Appendix A.2 and Equation 2 without proof; if the segmentation is not semantically coherent, the visual and text tokens lose the claimed meaning.
  • domain assumption OpenStreetMap provides complete and accurate road networks and POIs in tested cities.
    Visual tokens are rendered from OSM tiles; missing or inaccurate map features would degrade MLLM reasoning.
  • domain assumption The MLLM backends are treated as fixed, stable, and reproducible during inference and prompt optimization.
    Commercial APIs such as o4-mini and Gemini are not deterministic or version-pinned; prompt optimization may fit to transient model behavior.
  • ad hoc to paper The full trajectory, including all timestamps, can be used as input to TTE and MP without leaking the label.
    If the target time or destination is present in the input, the reported improvements are not evidence of generalization. The paper does not state that masking is applied.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Traj-MLLM: Can Multimodal Large Language Models Reform Trajectory Data Mining?." pith.science (2026). https://pith.science/paper/L3ETR4QV

@misc{pith2026250900053,
  author       = {Pith},
  title        = {Pith review of: Traj-MLLM: Can Multimodal Large Language Models Reform Trajectory Data Mining?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L3ETR4QV}},
  note         = {Machine review of arXiv:2509.00053}
}
abstract

Building a general model capable of analyzing human trajectories across different geographic regions and different tasks becomes an emergent yet important problem for various applications. However, existing works suffer from the generalization problem, \ie, they are either restricted to train for specific regions or only suitable for a few tasks. Given the recent advances of multimodal large language models (MLLMs), we raise the question: can MLLMs reform current trajectory data mining and solve the problem? Nevertheless, due to the modality gap of trajectory, how to generate task-independent multimodal trajectory representations and how to adapt flexibly to different tasks remain the foundational challenges. In this paper, we propose \texttt{Traj-MLLM}}, which is the first general framework using MLLMs for trajectory data mining. By integrating multiview contexts, \texttt{Traj-MLLM}} transforms raw trajectories into interleaved image-text sequences while preserving key spatial-temporal characteristics, and directly utilizes the reasoning ability of MLLMs for trajectory analysis. Additionally, a prompt optimization method is proposed to finalize data-invariant prompts for task adaptation. Extensive experiments on four publicly available datasets show that \texttt{Traj-MLLM}} outperforms state-of-the-art baselines by $48.05\%$, $15.52\%$, $51.52\%$, $1.83\%$ on travel time estimation, mobility prediction, anomaly detection and transportation mode identification, respectively. \texttt{Traj-MLLM}} achieves these superior performances without requiring any training data or fine-tuning the MLLM backbones.

Figures

Figures reproduced from arXiv: 2509.00053 by the authors.

Figure 1
Figure 1. The motivation of Traj-MLLM. potential to understand trajectories from visual and textual modal￾ities without extra training and extend to analyze trajectories in unseen regions. However, due to the modality gap of trajectory, two foundational challenges should be addressed to achieve region-agnostic and task￾adaptive trajectory data mining with MLLMs: • Task-independent Trajectory Modeling. The spatial-temporal cha… view at source ↗
Figure 2
Figure 2. Overview of Traj-MLLM. Instead of direct model training, Traj-MLLM transforms raw trajectories into interleaved image-text sequences through map-anchored tokenization and multiview trajectory modeling. This enables us to reframe downstream tasks as multimodal reasoning problems, which can be directly solved by MLLMs in a training-free manner. enhancer [41] and agents [5] to boost trajectory data mining. Nev￾ertheles… view at source ↗
Figure 3
Figure 3. Task Prompt Optimization. 4.3.1 Task Prompt Template. For each trajectory mining task 𝜏, we generate the task prompt with a general prompt template. The task prompt 𝑃sys consists of four components: 𝑃sys = (𝑃role, 𝑃task, 𝑃know, 𝑃format) (7) Here, the Role Definition (𝑃role) and Output Format (𝑃format) are fixed for each task. The optimization focuses on refining the Task Description (𝑃task) and Domain Knowledge (𝑃kn… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Performance with different MLLMs [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Sensitivity Analysis. structure favors a smaller 50m threshold to avoid interference from unrelated elements. More details are in Appendix A.4. 6 Conclusion In this paper, we propose a novel training-free framework Traj-MLLM which is the first work employing the reason…
Figure 6
Figure 6. Figure 6: Visual comparison of abnormal trajectories under [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: The result of Traj-MLLM [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: The result of Anomaly Detection. Ground Truth: Anomaly [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: The result of Transportation Mode Identification. Ground Truth: Train [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: The system prompt of Travel Time Estimation. [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: The result of Travel Time Estimation. Ground Truth: 2016-11-03 22:29:01 [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]
Figure 12
Figure 12. Figure 12: Qualitative comparison of outputs from different MLLM backbones for the Anomaly Detection task. The same input [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 37 canonical work pages

  1. [1]

    Yanchuan Chang, Jianzhong Qi, Yuxuan Liang, and Egemen Tanin. 2023. Con- trastive trajectory similarity learning with dual-feature attention. In 2023 IEEE 39th International conference on data engineering (ICDE) . IEEE, 2933–2945

    work page 2023

  2. [2]

    Yile Chen, Xiucheng Li, Gao Cong, Zhifeng Bao, Cheng Long, Yiding Liu, Arun Ku- mar Chandran, and Richard Ellison. 2021. Robust road network representation learning: When traffic patterns meet traveling semantics. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management . 211–220

    2021

  3. [3]

    Sina Dabiri, Chang-Tien Lu, Kevin Heaslip, and Chandan K Reddy. 2019. Semi- supervised deep learning approach for transportation mode identification using GPS trajectory data. IEEE Transactions on Knowledge and Data Engineering 32, 5 (2019), 1010–1023

    work page 2019

  4. [4]

    Yuhao Dong, Zuyan Liu, Hai-Long Sun, Jingkang Yang, Winston Hu, Yongming Rao, and Ziwei Liu. 2025. Insight-v: Exploring long-chain visual reasoning with multimodal large language models. In Proceedings of the Computer Vision and Pattern Recognition Conference. 9062–9072

    work page 2025

  5. [5]

    Yuwei Du, Jie Feng, Jie Zhao, and Yong Li. 2024. Trajagent: An agent framework for unified trajectory modelling. arXiv e-prints (2024), arXiv–2410

    work page 2024

  6. [6]

    Jie Feng, Yong Li, Chao Zhang, Funing Sun, Fanchao Meng, Ang Guo, and Depeng Jin. 2018. Deepmove: Predicting human mobility with attentional recurrent networks. In Proceedings of the 2018 world wide web conference . 1459–1468

    2018

  7. [7]

    Tao-Yang Fu and Wang-Chien Lee. 2020. Trembr: Exploring road networks for trajectory representation learning. ACM Transactions on Intelligent Systems and Technology (TIST) 11, 1 (2020), 1–25

    work page 2020

  8. [8]

    Qiang Gao, Xiaohan Wang, Chaoran Liu, Goce Trajcevski, Li Huang, and Fan Zhou. 2023. Open anomalous trajectory recognition via probabilistic metric learning. In IJCAI. International Joint Conferences on Artificial Intelligence Organization

    work page 2023

Show all 50 references
  1. [9]

    Xiaolin Han, Reynold Cheng, Chenhao Ma, and Tobias Grubenmann. 2022. DeepTEA: Effective and efficient online time-dependent trajectory outlier detec- tion. Proceedings of the VLDB Endowment 15, 7 (2022), 1493–1505

    2022

  2. [10]

    Jiawei Jiang, Dayan Pan, Houxing Ren, Xiaohan Jiang, Chao Li, and Jingyuan Wang. 2023. Self-supervised trajectory representation learning with temporal regularities and travel semantics. In 2023 IEEE 39th international conference on data engineering (ICDE). IEEE, 843–855

    2023

  3. [11]

    Xiang Jiang, Erico N de Souza, Ahmad Pesaranghader, Baifan Hu, Daniel L Silver, and Stan Matwin. 2017. Trajectorynet: An embedded gps trajectory representation for point-based classification using recurrent neural networks. arXiv preprint arXiv:1705.02636 (2017)

    2017 arXiv

  4. [12]

    Jihyung Kil, Zheda Mai, Justin Lee, Arpita Chowdhury, Zihe Wang, Kerrie Cheng, Lemeng Wang, Ye Liu, and Wei-Lun Harry Chao. 2024. Mllm-compbench: A comparative reasoning benchmark for multimodal llms. Advances in Neural Information Processing Systems 37 (2024), 28798–28827

    2024

  5. [13]

    Kartik Kuckreja, Muhammad Sohail Danish, Muzammal Naseer, Abhijit Das, Salman Khan, and Fahad Shahbaz Khan. 2024. Geochat: Grounded large vision- language model for remote sensing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 27831–27840

    2024

  6. [14]

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. 2023. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In International conference on machine learning . PMLR, 19730–19742

    2023

  7. [15]

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. 2022. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In International conference on machine learning . PMLR, 12888–12900

    2022

  8. [16]

    Ling Li, Yu Ye, Bingchuan Jiang, and Wei Zeng. 2024. Georeasoner: Geo- localization with reasoning in street views using a large vision-language model. In Forty-first International Conference on Machine Learning

    2024

  9. [17]

    Wenbin Li, Di Yao, Ruibo Zhao, Wenjie Chen, Zijie Xu, Chengxue Luo, Chang Gong, Quanliang Jing, Haining Tan, and Jingping Bi. 2025. Stbench: Assessing the ability of large language models in spatio-temporal analysis. In Companion Proceedings of the ACM on Web Conference 2025 . 749–752

    2025

  10. [18]

    Xiucheng Li, Kaiqi Zhao, Gao Cong, Christian S Jensen, and Wei Wei. 2018. Deep representation learning for trajectory similarity computation. In 2018 IEEE 34th international conference on data engineering (ICDE) . IEEE, 617–628

    2018

  11. [19]

    Yuxuan Liang, Kun Ouyang, Yiwei Wang, Xu Liu, Hongyang Chen, Junbo Zhang, Yu Zheng, and Roger Zimmermann. 2022. TrajFormer: Efficient trajectory classifi- cation with transformers. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management . ...

    2022

  12. [20]

    Yuxuan Liang, Kun Ouyang, Hanshu Yan, Yiwei Wang, Zekun Tong, and Roger Zimmermann. 2021. Modeling Trajectories with Neural Ordinary Differential Equations.. In IJCAI. 1498–1504

    2021

  13. [21]

    Yan Lin, Huaiyu Wan, Shengnan Guo, Jilin Hu, Christian S Jensen, and Youfang Lin. 2023. Pre-training general trajectory embeddings with maximum multi-view entropy coding. IEEE Transactions on Knowledge and Data Engineering 36, 12 (2023), 9037–9050

    2023

  14. [22]

    Yan Lin, Huaiyu Wan, Shengnan Guo, and Youfang Lin. 2021. Pre-training context and time aware location embeddings from spatial-temporal trajectories for user next location prediction. InProceedings of the AAAI conference on artificial intelligence, Vol. 35. 4241–4248

    2021

  15. [23]

    Yiding Liu, Kaiqi Zhao, Gao Cong, and Zhifeng Bao. 2020. Online anomalous trajectory detection with deep generative sequence modeling. In 2020 IEEE 36th International Conference on Data Engineering (ICDE) . IEEE, 949–960

    2020

  16. [24]

    Zhipeng Ma, Zheyan Tu, Xinhai Chen, Yan Zhang, Deguo Xia, Guyue Zhou, Yilun Chen, Yu Zheng, and Jiangtao Gong. 2024. More than routing: Joint GPS and route modeling for refine trajectory representation learning. In Proceedings of the ACM Web Conference 2024 . 3064–3075

    2024

  17. [25]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learni...

    2021

  18. [26]

    Wei Shao, Ziquan Fang, lu Chen, and yunjun Gao. 2025. Towards Trajectory Anomaly Detection: A Fine-Grained and Noise-Resilient Framework. In Pro- ceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    2025

  19. [27]

    Jiahui Sun, Haiming Jin, Zhaoxing Yang, Lu Su, and Xinbing Wang. 2022. Optimiz- ing long-term efficiency and fairness in ride-hailing via joint order dispatching and driver repositioning. In Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining ....

    2022

  20. [28]

    Vicente Vivanco Cepeda, Gaurav Kumar Nayak, and Mubarak Shah. 2023. Geoclip: Clip-inspired alignment between locations and images for effective worldwide geo-localization. Advances in Neural Information Processing Systems 36 (2023), 8690–8701

    2023

  21. [29]

    Chenhao Wang, Lisi Chen, Shuo Shang, Christian S Jensen, and Panos Kalnis

  22. [30]

    Yongfu Wei, Yan Lin, Hongfan Gao, Ronghui Xu, Sean Bin Yang, and Jilin Hu

  23. [31]

    Ronghui Xu, Hanyin Cheng, Chenjuan Guo, Hongfan Gao, Jilin Hu, Sean Bin Yang, and Bin Yang. 2025. Mm-path: Multi-modal, multi-granularity path representa- tion learning. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 1 . 1703–1714

    2025

  24. [32]

    Shixiong Xu, Chenghao Zhang, Lubin Fan, Gaofeng Meng, Shiming Xiang, and Jieping Ye. 2024. Addressclip: Empowering vision-language models for city-wide image address localization. In European Conference on Computer Vision . Springer, 76–92. Conference’17, July 2017, Washington...

    2024

  25. [33]

    Peilun Yang, Hanchen Wang, Ying Zhang, Lu Qin, Wenjie Zhang, and Xuemin Lin

  26. [34]

    Sean Bin Yang, Jilin Hu, Chenjuan Guo, Bin Yang, and Christian S Jensen. 2023. Lightpath: Lightweight and scalable path representation learning. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 2999–3010

    2023

  27. [35]

    Di Yao, Gao Cong, Chao Zhang, and Jingping Bi. 2019. Computing trajectory similarity in linear time: A generic seed-guided neural metric learning approach. In 2019 IEEE 35th international conference on data engineering (ICDE) . IEEE, 1358– 1369

    2019

  28. [36]

    Di Yao, Chao Zhang, Zhihua Zhu, Jianhui Huang, and Jingping Bi. 2017. Trajectory clustering via deep representation learning. In 2017 international joint conference on neural networks (IJCNN) . IEEE, 3880–3887

    2017

  29. [37]

    Xie Yu, Jingyuan Wang, Yifan Yang, Qian Huang, and Ke Qu. 2024. BIGCity: A universal spatiotemporal model for unified trajectory and traffic state data analysis. arXiv preprint arXiv:2412.00953 (2024)

    2024 arXiv

  30. [38]

    Daqing Zhang, Nan Li, Zhi-Hua Zhou, Chao Chen, Lin Sun, and Shijian Li. 2011. iBAT: detecting anomalous taxi trajectories from GPS traces. In Proceedings of the 13th international conference on Ubiquitous computing . 99–108

    2011

  31. [39]

    Hanyuan Zhang, Xingyu Zhang, Qize Jiang, Baihua Zheng, Zhenbang Sun, Wei- wei Sun, and Changhu Wang. 2020. Trajectory similarity learning with auxiliary supervision and optimal matching. (2020)

    2020

  32. [40]

    Wei Zhang, Miaoxin Cai, Tong Zhang, Yin Zhuang, and Xuerui Mao. 2024. Earth- GPT: A universal multimodal large language model for multisensor image com- prehension in remote sensing domain. IEEE Transactions on Geoscience and Remote Sensing 62 (2024), 1–20

    2024

  33. [41]

    Zheng Zhang, Hossein Amiri, Zhenke Liu, Liang Zhao, and Andreas Züfle. 2024. Large language models for spatial trajectory patterns mining. In Proceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection. 52–55

    2024

  34. [42]

    Jie Zhao, Chao Chen, Yuanshao Zhu, Mingyu Deng, and Yuxuan Liang. 2025. UniTR: A Unified Framework for Joint Representation Learning of Trajectories and Road Networks. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 13348–13356

    2025

  35. [43]

    Ge Zheng, Bin Yang, Jiajin Tang, Hong-Yu Zhou, and Sibei Yang. 2023. Ddcot: Duty-distinct chain-of-thought prompting for multimodal reasoning in language models. Advances in Neural Information Processing Systems 36 (2023), 5168–5191

    2023

  36. [44]

    Yu Zheng, Hao Fu, Xing Xie, Wei-Ying Ma, and Quannan Li. 2011. Geolife GPS Trajectory Dataset – User Guide . Technical Report. Microsoft Research Asia

    2011

  37. [45]

    Silin Zhou, Yao Chen, shuo Shang, lisi Chen, Bingsheng He, and Shibasaki Ryosuke. 2025. Blurred Encoding for Trajectory Representation Learning. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    2025

  38. [46]

    Silin Zhou, Shuo Shang, Lisi Chen, Christian S Jensen, and Panos Kalnis. 2024. RED: Effective Trajectory Representation Learning with Comprehensive Infor- mation. arXiv preprint arXiv:2411.15096 (2024)

    2024 arXiv

  39. [47]

    Final Answer

    Zeyu Zhou, Yan Lin, Haomin Wen, Shengnan Guo, Jilin Hu, Youfang Lin, and Huaiyu Wan. 2024. PLM4Traj: Cognizing Movement Patterns and Travel Pur- poses from Trajectories with Pre-trained Language Models. arXiv preprint arXiv:2405.12459 (2024). Traj-MLLM: Can Multimodal Large La...

    2024 arXiv

  40. [2021]

    In 2021 IEEE 37th international conference on data engineering (ICDE)

    T3s: Effective representation learning for trajectory similarity computation. In 2021 IEEE 37th international conference on data engineering (ICDE) . IEEE, 2183– 2188

    2021

  41. [2024]

    In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Multi-Scale Detection of Anomalous Spatio-Temporal Trajectories in Evolving Trajectory Datasets. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 2980–2990

  42. [2025]

    InProceedings of the ACM on Web Conference

    Path-LLM: A Multi-Modal Path Representation Learning by Aligning and Fusing with Large Language Models. InProceedings of the ACM on Web Conference

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.