pith. sign in

arxiv: 2511.12947 · v2 · submitted 2025-11-17 · 💻 cs.IR

ReST: A Plug-and-Play Spatially-Constrained Representation Enhancement Framework for Local-Life Recommendation

Hao Jiang , Long Zhang , Guoquan Wang , Sheng Yu , Yang Zeng , Wencong Zeng , Fei Pan , Peng Jiang
show 1 more author
Guorui Zhou
This is my paper

Pith reviewed 2026-05-17 21:30 UTC · model grok-4.3

classification 💻 cs.IR
keywords local-life recommendationlong-tail recommendationrepresentation learningspatial constraintscontrastive learningrecommendation framework
0
0 comments X

The pith

ReST enhances long-tail item representations in local-life recommendations by capturing latent relationships under spatial constraints.

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

The paper seeks to solve spatial constraints and long-tail sparsity in recommending local services like daily essentials. Existing user-centric methods are set aside in favor of an item-centric view that targets improving representations for less popular items. ReST starts by warming up ID embeddings with attribute information through a meta network. It then applies a contrastive learning module called SIDENet that uses hard sampling limited to spatial areas and dynamic alignment to strengthen weak representations. A sympathetic reader would care because this could make high-quality local options more visible despite their limited geographic reach.

Core claim

The central discovery is that a Meta ID Warm-up Network combined with the Spatially-Constrained ID Representation Enhancement Network can adaptively identify weak ID representations using attribute-level information and enhance them by capturing latent item relationships that respect the spatial constraints of local lifestyle services, all while preserving performance on popular items.

What carries the argument

SIDENet, a contrastive learning network that incorporates spatially-constrained hard sampling and dynamic representation alignment to enhance weak ID representations.

If this is right

  • Enhanced representations for long-tail items lead to better recommendations in geographically limited areas.
  • The approach maintains compatibility with popular items during the enhancement process.
  • Being plug-and-play allows integration with various existing recommendation models.
  • Focus on item relationships within spatial constraints addresses the exposure imbalance in local-life scenarios.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Applying similar spatially constrained sampling could benefit recommendation in other domains like ride-sharing or local events where location limits exposure.
  • Future tests might examine performance when spatial areas vary in size or density to see robustness.
  • The method's reliance on attribute information suggests it could extend to scenarios with rich item metadata.

Load-bearing premise

That shifting to an item-centric perspective and using the proposed sampling and alignment strategies will reliably boost weak representations without causing negative effects on the overall system.

What would settle it

Observing no improvement or even a decline in recommendation accuracy for long-tail items in real local-life datasets when using ReST compared to standard models would indicate the claim does not hold.

Figures

Figures reproduced from arXiv: 2511.12947 by Fei Pan, Guoquan Wang, Guorui Zhou, Hao Jiang, Long Zhang, Peng Jiang, Sheng Yu, Wencong Zeng, Yang Zeng.

Figure 1
Figure 1. Figure 1: Illustration of local lifestyle recommendation: spa [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: This figure illustrates the overall structure and key components of the ReST model. Specifically:(a) shows the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Ablation study on ReST. As shown in [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance comparison with different numbers of hard negative samples. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance improvement in orders and GMV [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
read the original abstract

Local-life recommendation have witnessed rapid growth, providing users with convenient access to daily essentials. However, this domain faces two key challenges: (1) spatial constraints, driven by the requirements of the local-life scenario, where items are usually shown only to users within a limited geographic area, indirectly reducing their exposure probability; and (2) long-tail sparsity, where few popular items dominate user interactions, while many high-quality long-tail items are largely overlooked due to imbalanced interaction opportunities. Existing methods typically adopt a user-centric perspective, such as modeling spatial user preferences or enhancing long-tail representations with collaborative filtering signals. However, we argue that an item-centric perspective is more suitable for this domain, focusing on enhancing long-tail items representation that align with the spatially-constrained characteristics of local lifestyle services. To tackle this issue, we propose ReST, a Plug-And-Play Spatially-Constrained Representation Enhancement Framework for Long-Tail Local-Life Recommendation. Specifically, we first introduce a Meta ID Warm-up Network, which initializes fundamental ID representations by injecting their basic attribute-level semantic information. Subsequently, we propose a novel Spatially-Constrained ID Representation Enhancement Network (SIDENet) based on contrastive learning, which incorporates two efficient strategies: a spatially-constrained hard sampling strategy and a dynamic representation alignment strategy. This design adaptively identifies weak ID representations based on their attribute-level information during training. It additionally enhances them by capturing latent item relationships within the spatially-constrained characteristics of local lifestyle services, while preserving compatibility with popular items.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The paper proposes ReST, a plug-and-play framework for long-tail local-life recommendation that adopts an item-centric perspective. It introduces a Meta ID Warm-up Network to initialize ID embeddings with attribute-level semantics, followed by a Spatially-Constrained ID Representation Enhancement Network (SIDENet) that uses contrastive learning with a spatially-constrained hard sampling strategy and a dynamic representation alignment strategy to strengthen weak long-tail item representations while preserving performance on popular items. The approach targets spatial constraints and interaction sparsity typical of local-life services.

Significance. If the reported empirical gains and ablations hold, ReST offers a practical, modular enhancement for recommendation models in spatially constrained domains. The item-centric design, combined with explicit handling of long-tail items without negative transfer to head items, addresses a relevant gap; the plug-and-play integration and ablation studies on sampling and alignment strategies add to its potential utility.

major comments (2)
  1. [§3.3] §3.3 (SIDENet description): the dynamic representation alignment strategy is presented as adaptively identifying weak representations, but the precise mechanism for computing the alignment loss weight or the threshold for 'weak' status based on attribute information is not formalized; this leaves open whether the adaptivity is fully parameter-free or requires dataset-specific tuning that could affect the central claim of reliable enhancement.
  2. [§4.3] §4.3 (experimental setup): while ablation results on long-tail metrics are reported, the paper does not include a direct comparison against recent user-centric spatial preference models (e.g., those using geographic embeddings) on the same local-life datasets; this weakens the justification for preferring the item-centric approach over established alternatives.
minor comments (3)
  1. [Abstract] Abstract: 'Local-life recommendation have witnessed' contains a subject-verb agreement error and should read 'has witnessed'.
  2. [§2] §2 (related work): the discussion of existing user-centric methods could more explicitly cite recent contrastive learning approaches for long-tail recommendation to better position the novelty of the spatially-constrained sampling.
  3. [Figure 2] Figure 2 (framework diagram): the flow from Meta ID Warm-up to SIDENet is clear, but the notation for the contrastive loss components could be labeled more explicitly to match the equations in §3.2.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive evaluation and constructive comments, which help improve the clarity and justification of our work. We address each major comment point by point below, indicating the revisions planned for the manuscript.

read point-by-point responses

  1. Referee: [§3.3] §3.3 (SIDENet description): the dynamic representation alignment strategy is presented as adaptively identifying weak representations, but the precise mechanism for computing the alignment loss weight or the threshold for 'weak' status based on attribute information is not formalized; this leaves open whether the adaptivity is fully parameter-free or requires dataset-specific tuning that could affect the central claim of reliable enhancement.

    Authors: We thank the referee for this observation on the need for formalization. The current manuscript describes the dynamic representation alignment strategy at a conceptual level, noting that it adaptively identifies weak ID representations based on attribute-level information during training. To address the concern, we will revise §3.3 to include an explicit mathematical formulation for the alignment loss weight and the threshold used to determine 'weak' status. This will clarify the computation process and demonstrate that the mechanism is designed to function adaptively based on attribute information without requiring extensive dataset-specific tuning, thereby reinforcing the reliability of the enhancement for long-tail items. revision: yes

  2. Referee: [§4.3] §4.3 (experimental setup): while ablation results on long-tail metrics are reported, the paper does not include a direct comparison against recent user-centric spatial preference models (e.g., those using geographic embeddings) on the same local-life datasets; this weakens the justification for preferring the item-centric approach over established alternatives.

    Authors: We appreciate the referee's suggestion to strengthen the experimental justification. Our work specifically advocates an item-centric perspective because local-life recommendations are governed by item-level spatial constraints and exposure limitations, which differ from user-centric modeling of geographic preferences. To better support this choice, we will add a dedicated discussion paragraph in the revised manuscript (likely in §2 or §4) that contrasts the item-centric design with representative user-centric spatial models, highlighting domain-specific advantages without negative transfer to head items. While we do not plan new full-scale experiments with those models for this minor revision, the expanded discussion will provide clearer conceptual justification. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper introduces ReST as a new plug-and-play framework with a Meta ID Warm-up Network for attribute-level initialization and SIDENet using contrastive learning with spatially-constrained hard sampling plus dynamic alignment. No equations or steps in the provided description reduce the output representations or predictions directly to fitted parameters or self-referential definitions by construction. The item-centric premise is motivated by domain-specific spatial constraints and long-tail issues rather than uniqueness theorems imported from self-citations. Ablation results and plug-and-play integration details supply independent empirical content, keeping the central claims from collapsing into input fitting or renaming of known patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Paper introduces two new network modules and two training strategies whose effectiveness is asserted without independent derivation or external benchmarks visible in the abstract.

invented entities (2)
  • Meta ID Warm-up Network no independent evidence
    purpose: Initializes fundamental ID representations by injecting basic attribute-level semantic information
    New component introduced to address cold-start or sparse ID issues
  • Spatially-Constrained ID Representation Enhancement Network (SIDENet) no independent evidence
    purpose: Enhances weak long-tail ID representations via contrastive learning with spatial constraints
    Core novel network proposed for the local-life setting

pith-pipeline@v0.9.0 · 5604 in / 1356 out tokens · 38272 ms · 2026-05-17T21:30:21.979427+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Birds of a Feather Cluster Nearby: a Proximity-Aware Geo-Codebook for Local Service Recommendation

    cs.IR 2026-04 unverdicted novelty 6.0

    Pro-GEO introduces a geo-centroid coordinate system and geo-rotary position encoding to model geographic proximity as rotational transformations, enabling balanced semantic-spatial modeling in local service recommendations.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages · cited by 1 Pith paper · 3 internal anchors

  1. [1]

    Representation Learning with Contrastive Predictive Coding

    A. v. d. Oord, Y. Li, and O. Vinyals, “Representation learning with contrastive predictive coding, ”arXiv preprint arXiv:1807.03748, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  2. [2]

    User modeling with click preference and reading satisfaction for news recommendation,

    C. Wu, F. Wu, T. Qi, and Y. Huang, “User modeling with click preference and reading satisfaction for news recommendation, ” inProceedings of the Twenty- Ninth International Joint Conference on Artificial Intelligence, ser. IJCAI’20, 2021

    work page 2021

  3. [3]

    Knowledge graph unlearning to defend language model against jailbreak attack,

    P. Mai, H. Jiang, R. Yan, Y. Yang, Z. Huang, and Y. Pang, “Knowledge graph unlearning to defend language model against jailbreak attack, ” in The Second Tiny Papers Track at ICLR 2024, 2024. [Online]. Available: https://openreview.net/forum?id=gqTUIesy4H

    work page 2024

  4. [5]

    Spatiotemporal-enhanced network for click-through rate prediction in location- based services,

    S. Lin, Y. Yu, X. Ji, T. Zhou, H. He, Z. Sang, J. Jia, G. Cao, and N. Hu, “Spatiotemporal-enhanced network for click-through rate prediction in location- based services, ”arXiv preprint arXiv:2209.09427, 2022

    work page arXiv 2022

  5. [6]

    Improving long-tail item recommendation with graph augmentation,

    S. Luo, C. Ma, Y. Xiao, and L. Song, “Improving long-tail item recommendation with graph augmentation, ” inProceedings of the 32nd ACM international conference on information and knowledge management, 2023, pp. 1707–1716

    work page 2023

  6. [8]

    Empowering long-tail item recommendation through cross decoupling network (cdn),

    Y. Zhang, R. Wang, D. Z. Cheng, T. Yao, X. Yi, L. Hong, J. Caverlee, and E. H. Chi, “Empowering long-tail item recommendation through cross decoupling network (cdn), ” inProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023, pp. 5608–5617

    work page 2023

  7. [10]

    Session-based Recommendations with Recurrent Neural Networks

    B. Hidasi, A. Karatzoglou, L. Baltrunas, and D. Tikk, “Session-based recommen- dations with recurrent neural networks, ”arXiv preprint arXiv:1511.06939, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  8. [11]

    Enhancing sequential recommendation with graph contrastive learning,

    Y. Zhang, Y. Liu, Y. Xu, H. Xiong, C. Lei, W. He, L. Cui, and C. Miao, “Enhancing sequential recommendation with graph contrastive learning, ”arXiv preprint arXiv:2205.14837, 2022

    work page arXiv 2022

  9. [12]

    Building and exploiting spatial–temporal knowledge graph for next poi recommendation,

    W. Chen, H. Wan, S. Guo, H. Huang, S. Zheng, J. Li, S. Lin, and Y. Lin, “Building and exploiting spatial–temporal knowledge graph for next poi recommendation, ” Knowledge-Based Systems, vol. 258, p. 109951, 2022

    work page 2022

  10. [13]

    An attentive inductive bias for sequential recommendation beyond the self-attention,

    Y. Shin, J. Choi, H. Wi, and N. Park, “An attentive inductive bias for sequential recommendation beyond the self-attention, ” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 8, 2024, pp. 8984–8992

    work page 2024

  11. [14]

    Distillation from het- erogeneous models for top-k recommendation,

    S. Kang, W. Kweon, D. Lee, J. Lian, X. Xie, and H. Yu, “Distillation from het- erogeneous models for top-k recommendation, ” inProceedings of the ACM Web Conference 2023, 2023, pp. 801–811

    work page 2023

  12. [18]

    E-commerce recommendation applica- tions,

    J. B. Schafer, J. A. Konstan, and J. Riedl, “E-commerce recommendation applica- tions, ”Data mining and knowledge discovery, vol. 5, pp. 115–153, 2001

    work page 2001

  13. [19]

    Music recommendation,

    O. Celma, “Music recommendation, ” inMusic recommendation and discovery: The long tail, long fail, and long play in the digital music space. Springer, 2010, pp. 43–85

    work page 2010

  14. [20]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, S. Albanie, G. Sun, and E. Wu, “Squeeze-and-excitation networks, ”

    work page

  15. [21]

    Squeeze-and-Excitation Networks

    [Online]. Available: https://arxiv.org/abs/1709.01507

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [22]

    St-pil: Spatial-temporal periodic interest learning for next point-of-interest recommendation,

    Q. Cui, C. Zhang, Y. Zhang, J. Wang, and M. Cai, “St-pil: Spatial-temporal periodic interest learning for next point-of-interest recommendation, ” inProceedings of the 30th ACM International conference on information & knowledge management, 2021, pp. 2960–2964

    work page 2021

  17. [23]

    Trilateral spatiotemporal attention network for user behavior modeling in location-based search,

    Y. Qi, K. Hu, B. Zhang, J. Cheng, and J. Lei, “Trilateral spatiotemporal attention network for user behavior modeling in location-based search, ” inProceedings of the 30th ACM International Conference on Information & Knowledge Management, 2021, pp. 3373–3377

    work page 2021

  18. [25]

    Exploring the spatiotemporal features of online food recommendation service,

    S. Lin, J. Pei, T. Zhou, H. He, J. Jia, and N. Hu, “Exploring the spatiotemporal features of online food recommendation service, ” inProceedings of the 46th In- ternational ACM SIGIR Conference on Research and Development in Information Retrieval, 2023, pp. 3354–3358

    work page 2023

  19. [26]

    Basm: A bottom-up adaptive spatiotemporal model for online food ordering service,

    B. Du, S. Lin, J. Gao, X. Ji, M. Wang, T. Zhou, H. He, J. Jia, and N. Hu, “Basm: A bottom-up adaptive spatiotemporal model for online food ordering service, ” in 2023 IEEE 39th International Conference on Data Engineering (ICDE). IEEE, 2023, pp. 3549–3562

    work page 2023

  20. [27]

    Clsprec: Contrastive learning of long and short-term preferences for next poi recommendation,

    C. Duan, W. Fan, W. Zhou, H. Liu, and J. Wen, “Clsprec: Contrastive learning of long and short-term preferences for next poi recommendation, ” inProceedings of the 32nd ACM international conference on information and knowledge management, 2023, pp. 473–482

    work page 2023

  21. [28]

    Mcn4rec: Multi-level col- laborative neural network for next location recommendation,

    S. Li, W. Chen, B. Wang, C. Huang, Y. Yu, and J. Dong, “Mcn4rec: Multi-level col- laborative neural network for next location recommendation, ”ACM Transactions on Information Systems, vol. 42, no. 4, pp. 1–26, 2024

    work page 2024

  22. [29]

    Fragment and integrate network (fin): A novel spatial- temporal modeling based on long sequential behavior for online food ordering click-through rate prediction,

    J. Li and G. Zhang, “Fragment and integrate network (fin): A novel spatial- temporal modeling based on long sequential behavior for online food ordering click-through rate prediction, ” inProceedings of the 32nd ACM International Conference on Information and Knowledge Management, 2023, pp. 4688–4694

    work page 2023

  23. [30]

    Dropoutnet: Addressing cold start in rec- ommender systems,

    M. Volkovs, G. Yu, and T. Poutanen, “Dropoutnet: Addressing cold start in rec- ommender systems, ”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  24. [31]

    Warm up cold-start advertisements: Improving ctr predictions via learning to learn id embeddings,

    F. Pan, S. Li, X. Ao, P. Tang, and Q. He, “Warm up cold-start advertisements: Improving ctr predictions via learning to learn id embeddings, ” inProceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval, 2019, pp. 695–704

    work page 2019

  25. [32]

    Towards long-term fairness in recommendation,

    Y. Ge, S. Liu, R. Gao, Y. Xian, Y. Li, X. Zhao, C. Pei, F. Sun, J. Ge, W. Ouet al., “Towards long-term fairness in recommendation, ” inProceedings of the 14th ACM international conference on web search and data mining, 2021, pp. 445–453

    work page 2021

  26. [33]

    Contrastive learning for sequential recommendation,

    X. Xie, F. Sun, Z. Liu, S. Wu, J. Gao, J. Zhang, B. Ding, and B. Cui, “Contrastive learning for sequential recommendation, ” in2022 IEEE 38th international confer- ence on data engineering (ICDE). IEEE, 2022, pp. 1259–1273

    work page 2022

  27. [34]

    Alleviating the long-tail problem in conversational recommender systems,

    Z. Zhao, K. Zhou, X. Wang, W. X. Zhao, F. Pan, Z. Cao, and J.-R. Wen, “Alleviating the long-tail problem in conversational recommender systems, ” inProceedings of the 17th ACM Conference on Recommender Systems, 2023, pp. 374–385

    work page 2023

  28. [35]

    Hierarchical interest modeling of long-tailed users for click-through rate pre- diction,

    X. Xie, J. Niu, L. Deng, D. Wang, J. Zhang, Z. Wu, K. Bian, G. Cao, and B. Cui, “Hierarchical interest modeling of long-tailed users for click-through rate pre- diction, ” in2023 IEEE 39th International Conference on Data Engineering (ICDE). IEEE, 2023, pp. 3058–3071

    work page 2023

  29. [36]

    Biased user history synthesis for personalized long-tail item recom- mendation,

    K. Balasubramanian, A. Alshabanah, E. Markowitz, G. Ver Steeg, and M. An- navaram, “Biased user history synthesis for personalized long-tail item recom- mendation, ” inProceedings of the 18th ACM Conference on Recommender Systems, 2024, pp. 189–199

    work page 2024

  30. [37]

    Ppm: A pre-trained plug-in model for click-through rate prediction,

    Y. Gao, P. Lin, D. Wang, F. Mei, X. Zhao, S. Xu, and J. Hu, “Ppm: A pre-trained plug-in model for click-through rate prediction, ” inCompanion Proceedings of the ACM Web Conference 2024, 2024, pp. 311–318

    work page 2024

  31. [38]

    Post-training embedding enhancement for long-tail recommendation,

    G. Lee, K. Kim, and K. Shin, “Post-training embedding enhancement for long-tail recommendation, ” inProceedings of the 33rd ACM International Conference on Information and Knowledge Management, 2024, pp. 3857–3861

    work page 2024

  32. [39]

    Temporal linear item-item model for sequential recommendation,

    S. Park, M. Yoon, M. Choi, and J. Lee, “Temporal linear item-item model for sequential recommendation, ” inProceedings of the Eighteenth ACM International Conference on Web Search and Data Mining, 2025, pp. 354–362. A Plug-and-Play Spatially-Constrained Representation Enhancement Framework for Local-Life Recommendation Conference’17, July 2017, Washington...

    work page 2025

  33. [40]

    Deep interest network for click-through rate prediction,

    G. Zhou, X. Zhu, C. Song, Y. Fan, H. Zhu, X. Ma, Y. Yan, J. Jin, H. Li, and K. Gai, “Deep interest network for click-through rate prediction, ” inProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, 2018, pp. 1059–1068

    work page 2018

  34. [41]

    Relative contrastive learning for sequential recommendation with similarity-based positive sample selection,

    Z. Wang, Y. Shen, Z. Zhang, L. He, Y. Li, H. Gu, and Y. Zhang, “Relative contrastive learning for sequential recommendation with similarity-based positive sample selection, ” inProceedings of the 33rd ACM International Conference on Information and Knowledge Management, 2024, pp. 2493–2502

    work page 2024

  35. [42]

    Spatial-temporal knowledge distillation for takeaway recommendation,

    S. Zhao, W. Chen, B. Shi, L. Zhou, S. Lin, and H. Wan, “Spatial-temporal knowledge distillation for takeaway recommendation, ” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 12, 2025, pp. 13 365–13 373

    work page 2025

  36. [44]

    The meaning and use of the area under a receiver operating characteristic (roc) curve

    J. A. Hanley and B. J. McNeil, “The meaning and use of the area under a receiver operating characteristic (roc) curve. ”Radiology, vol. 143, no. 1, pp. 29–36, 1982

    work page 1982

  37. [45]

    The trec-8 question answering track report

    E. M. Voorheeset al., “The trec-8 question answering track report. ” inTrec, vol. 99, 1999, pp. 77–82

    work page 1999

  38. [46]

    Cumulated gain-based evaluation of ir techniques,

    K. Järvelin and J. Kekäläinen, “Cumulated gain-based evaluation of ir techniques, ” ACM Transactions on Information Systems (TOIS), vol. 20, no. 4, pp. 422–446, 2002

    work page 2002

  39. [47]

    Neural collaborative filtering,

    X. He, L. Liao, H. Zhang, L. Nie, X. Hu, and T.-S. Chua, “Neural collaborative filtering, ” inProceedings of the 26th international conference on world wide web, 2017, pp. 173–182

    work page 2017

  40. [48]

    Qarm: Quantitative alignment multi-modal recommenda- tion at kuaishou.arXiv preprint arXiv:2411.11739,

    X. Luo, J. Cao, T. Sun, J. Yu, R. Huang, W. Yuan, H. Lin, Y. Zheng, S. Wang, Q. Hu et al., “Qarm: Quantitative alignment multi-modal recommendation at kuaishou, ” arXiv preprint arXiv:2411.11739, 2024

    work page arXiv 2024

  41. [49]

    The cosine-haversine formula,

    C. C. Robusto, “The cosine-haversine formula, ”The American Mathematical Monthly, vol. 64, no. 1, pp. 38–40, 1957

    work page 1957

  42. [50]

    Fim: Frequency- aware multi-view interest modeling for local-life service recommendation,

    G. Wang, Q. Luo, W. Hu, P. Yao, W. Zeng, G. Zhou, and K. Gai, “Fim: Frequency- aware multi-view interest modeling for local-life service recommendation, ”arXiv preprint arXiv:2504.17814, 2025

    work page arXiv 2025

  43. [51]

    Modeling spatiotem- poral periodicity and collaborative signal for local-life service recommendation,

    H. Chi, H. Xu, M. Liu, Y. Bei, S. Zhou, D. Liu, and M. Zhang, “Modeling spatiotem- poral periodicity and collaborative signal for local-life service recommendation, ” arXiv preprint arXiv:2309.12565, 2023

    work page arXiv 2023

  44. [52]

    Geo-sage: A geographical sparse additive generative model for spatial item recommendation,

    W. Wang, H. Yin, L. Chen, Y. Sun, S. Sadiq, and X. Zhou, “Geo-sage: A geographical sparse additive generative model for spatial item recommendation, ” inProceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining, 2015, pp. 1255–1264

    work page 2015

  45. [53]

    Stcaplrs: A spatial-temporal context-aware personalized location recommendation system,

    Q. Fang, C. Xu, M. S. Hossain, and G. Muhammad, “Stcaplrs: A spatial-temporal context-aware personalized location recommendation system, ”ACM Transactions on Intelligent systems and technology (TIST), vol. 7, no. 4, pp. 1–30, 2016

    work page 2016

  46. [54]

    Deep learning,

    Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning, ”nature, vol. 521, no. 7553, pp. 436–444, 2015

    work page 2015

  47. [55]

    A geohash-based index for spatial data management in distributed memory,

    J. Liu, H. Li, Y. Gao, H. Yu, and D. Jiang, “A geohash-based index for spatial data management in distributed memory, ” in2014 22Nd international conference on geoinformatics. IEEE, 2014, pp. 1–4

    work page 2014

  48. [56]

    k-means++: The advantages of careful seeding,

    D. Arthur and S. Vassilvitskii, “k-means++: The advantages of careful seeding, ” Stanford, Tech. Rep., 2006

    work page 2006