Generative Spatiotemporal Intent Sequence Recommendation via Implicit Reasoning in Amap

Bowen Zheng; Fanyi Di; Jie Li; Jun Meng; Ruiting Dong; Shuaijun Guo; Sicong Wang; Xin Li; Yue Liu; Yu Gu

arxiv: 2605.28888 · v1 · pith:2NPGS6WQnew · submitted 2026-05-27 · 💻 cs.IR · cs.LG

Generative Spatiotemporal Intent Sequence Recommendation via Implicit Reasoning in Amap

Sicong Wang , Ruiting Dong , Yue Liu , Bowen Zheng , Jun Meng , Jie Li , Shuaijun Guo , Yu Gu

show 2 more authors

Fanyi Di Xin Li

This is my paper

Pith reviewed 2026-06-29 10:07 UTC · model grok-4.3

classification 💻 cs.IR cs.LG

keywords generative spatiotemporal recommendationintent sequence generationimplicit CoT distillationcounterfactual DPOlightweight modelsLLM reasoning compressioncontext responsivenesssequence recommendation

0 comments

The pith

GPlan distills LLM reasoning into latent tokens so lightweight models generate coherent spatiotemporal intent sequences.

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

The paper tries to establish that explicit reasoning from large language models can be compressed into compact models to generate sequences of user intents that remain logically connected and physically possible under real time and location constraints, as in mapping services. Direct LLM use creates high latency and often yields plans that ignore actual spatiotemporal limits. The GPlan framework achieves this through two steps: progressive implicit chain-of-thought distillation that packs reasoning into reserved latent tokens, and spatiotemporal counterfactual DPO that trains the model on pairs highlighting context mismatches. A sympathetic reader would care because successful compression would enable real-time, executable recommendations in production systems without the cost or delay of full-scale language models.

Core claim

The paper claims that Progressive Implicit CoT Distillation compresses explicit LLM reasoning processes into reserved latent tokens so small models inherit complex planning logic, while Spatiotemporal Counterfactual DPO aligns the model on counterfactual context-plan pairs to heighten sensitivity to spatiotemporal constraints, together allowing generative models to produce intent sequences that are more coherent and context-responsive than direct LLM outputs or prior methods, as measured in offline experiments and online A/B tests on the GSISR task.

What carries the argument

The GPlan framework, built on Progressive Implicit CoT Distillation that packs reasoning into latent tokens and Spatiotemporal Counterfactual DPO that trains on counterfactual context-plan pairs.

If this is right

Lightweight models can replace full LLMs for intent sequence generation while meeting strict latency limits.
Generated sequences exhibit higher logical coherence and physical executability within given spatiotemporal contexts.
Fewer plans mismatch the actual time and location constraints of the user.
The approach supports practical deployment in industrial mapping and service recommendation systems.
The released GSISR dataset allows direct replication and extension of the method.

Where Pith is reading between the lines

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

The same latent-token compression could transfer to other sequential planning tasks that involve physical constraints, such as delivery routing or urban mobility suggestions.
If the reserved tokens capture general planning patterns, the distillation step might reduce reliance on LLMs across broader recommendation domains.
Online A/B gains suggest the method could raise overall user retention when intent sequences are presented in live map applications.
Combining the counterfactual alignment with additional efficiency techniques like quantization might yield further latency reductions without losing coherence.

Load-bearing premise

Compressing explicit LLM reasoning into reserved latent tokens preserves the full planning logic, and counterfactual alignment on context-plan pairs is enough to bridge general knowledge to real-world spatiotemporal constraints.

What would settle it

If small models trained with GPlan show no gains in sequence coherence metrics or fail to improve user engagement metrics in A/B tests compared to baselines that skip the latent-token distillation and counterfactual DPO steps, the central claim would not hold.

Figures

Figures reproduced from arXiv: 2605.28888 by Bowen Zheng, Fanyi Di, Jie Li, Jun Meng, Ruiting Dong, Shuaijun Guo, Sicong Wang, Xin Li, Yue Liu, Yu Gu.

**Figure 1.** Figure 1: The GPlan training and inference pipeline. (Top) A teacher generates a structured CoT and a JSON intent sequence; [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

read the original abstract

Real-world user behavior rarely consists of isolated actions; instead, it often forms intent flows governed by spatiotemporal dependencies. To provide integrated service recommendations, we focus on the task of Generative Spatiotemporal Intent Sequence Recommendation (GSISR), which aims to generate intent sequences that are logically coherent and physically executable within complex spatiotemporal contexts. While LLMs offer strong reasoning potential for GSISR, direct industrial deployment is limited by high inference latency and context-mismatched or physically infeasible plans. To address these challenges, we propose a generative framework, GPlan, that internalizes LLM reasoning into lightweight models through two components. First, to enable reasoning under strict latency constraints, we introduce Progressive Implicit CoT Distillation, which compresses explicit reasoning processes into reserved latent tokens, allowing small models to inherit complex planning logic without generating long reasoning text. Second, to address the disconnect between general knowledge and real-world constraints, we design Spatiotemporal Counterfactual DPO. By aligning the model with counterfactual context-plan pairs, we improve sensitivity to spatiotemporal context and reduce context-mismatched plans. Offline experiments and online A/B testing demonstrate that our approach improves sequence coherence and context responsiveness. Our implementation and the anonymized GSISR dataset are available at https://github.com/alibaba/GPlan.

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.

Desk Editor's Note private letter to a colleague

GPlan distills LLM reasoning into latent tokens for low-latency spatiotemporal sequence recs, but the abstract gives no numbers or ablations so the gains are hard to evaluate.

read the letter

The main thing here is a practical framework called GPlan that tries to take LLM-style planning for intent sequences in location services and squeeze it into small models via two tricks: progressive implicit chain-of-thought distillation into reserved latent tokens, and a spatiotemporal counterfactual DPO step to better match real constraints. That combination looks new for this exact setting, even if distillation and DPO are not. The paper also ships code and an anonymized dataset, which is useful for anyone who wants to test it.

What it does reasonably well is target real deployment pain points—high latency and plans that ignore physical or temporal feasibility in apps like Amap. The online A/B mention suggests they ran it in production, which adds some weight.

The soft spot is the lack of concrete evidence in the abstract. No metrics, no baseline comparisons, no ablation isolating whether the latent-token compression actually keeps the multi-step logic intact versus just correlating with better scores on their test set. The stress-test concern about whether the distillation preserves complex spatiotemporal constraints is fair; without equations or direct comparisons in the summary, it is difficult to tell if the central mechanism works as claimed or if gains come from other factors. The full paper might have those details, but based on what is visible the empirical support stays thin.

This is for people building recommender systems in mobile location services who care about efficiency and coherence. It is not reshaping core theory. I would send it to peer review because the applied angle and released artifacts make it worth a closer look by referees who can check the experiments, even if heavy revision is likely.

Referee Report

2 major / 0 minor

Summary. The manuscript introduces GPlan, a generative framework for the task of Generative Spatiotemporal Intent Sequence Recommendation (GSISR). It proposes two components: Progressive Implicit CoT Distillation, which compresses explicit LLM reasoning processes into reserved latent tokens so that lightweight models can inherit planning logic without emitting long text, and Spatiotemporal Counterfactual DPO, which aligns the model on counterfactual context-plan pairs to improve sensitivity to real-world spatiotemporal constraints. The authors claim that offline experiments and online A/B testing show improvements in sequence coherence and context responsiveness, and they release code and an anonymized dataset.

Significance. If the empirical claims hold after proper verification, the work could enable practical deployment of complex reasoning in latency-constrained industrial recommendation systems for spatiotemporal services. The open release of implementation and dataset supports reproducibility and is a clear strength.

major comments (2)

[Abstract] Abstract: the central claim that the two components deliver measurable gains in coherence and responsiveness is asserted without any reported metrics, baselines, dataset statistics, ablation results, or quantitative details, so the data cannot be checked against the claim.
[Progressive Implicit CoT Distillation] Progressive Implicit CoT Distillation (as described in the abstract): the assertion that compression into latent tokens preserves complex multi-step spatiotemporal planning logic (route feasibility, time windows, physical executability) lacks any direct comparison, information-theoretic argument, or ablation isolating whether the representation is sufficient rather than merely correlated with metrics on the test distribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment below and indicate planned revisions.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that the two components deliver measurable gains in coherence and responsiveness is asserted without any reported metrics, baselines, dataset statistics, ablation results, or quantitative details, so the data cannot be checked against the claim.

Authors: The abstract is a concise summary and therefore omits specific numbers. The full manuscript (Section 4) reports the requested details: coherence and responsiveness metrics, baselines, dataset statistics, and ablation results from both offline experiments and online A/B tests. To improve verifiability, we will add the key quantitative gains to the abstract in the revision. revision: yes
Referee: [Progressive Implicit CoT Distillation] Progressive Implicit CoT Distillation (as described in the abstract): the assertion that compression into latent tokens preserves complex multi-step spatiotemporal planning logic (route feasibility, time windows, physical executability) lacks any direct comparison, information-theoretic argument, or ablation isolating whether the representation is sufficient rather than merely correlated with metrics on the test distribution.

Authors: Section 3 describes the distillation procedure and Section 4.3 presents ablations that isolate the component's contribution to coherence metrics tied to planning quality. These results provide empirical evidence that the latent tokens carry the necessary logic. We acknowledge the absence of an explicit information-theoretic bound or probing study; we will add a short discussion of representation sufficiency and one additional ablation in the revision. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical framework validated by offline and online tests

full rationale

The paper proposes the GPlan framework consisting of Progressive Implicit CoT Distillation and Spatiotemporal Counterfactual DPO to internalize LLM reasoning into lightweight models for GSISR. Central claims rest on empirical results from offline experiments and online A/B testing showing gains in coherence and responsiveness. No derivation chain, equations, or predictions that reduce by construction to fitted inputs or self-citations appear in the provided text. The work is self-contained against external benchmarks via direct testing rather than self-referential definitions or renamed known results.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the methods are described at the level of high-level components without equations or implementation details.

pith-pipeline@v0.9.1-grok · 5779 in / 1081 out tokens · 26007 ms · 2026-06-29T10:07:49.789819+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

29 extracted references · 20 canonical work pages · 5 internal anchors

[1]

Tesfaye Fenta Boka, Zhendong Niu, and Rama Bastola Neupane. 2024. A sur- vey of sequential recommendation systems: Techniques, evaluation, and future directions.Information Systems125 (2024), 102427. doi:10.1016/j.is.2024.102427

work page doi:10.1016/j.is.2024.102427 2024
[2]

Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin...

work page internal anchor Pith review Pith/arXiv arXiv 2020
[3]

Jeffrey Cheng and Benjamin Van Durme. 2024. Compressed Chain of Thought: Efficient Reasoning Through Dense Representations. arXiv:2412.13171 [cs.CL] https://arxiv.org/abs/2412.13171

work page internal anchor Pith review Pith/arXiv arXiv 2024
[4]

Yuntian Deng, Yejin Choi, and Stuart Shieber. 2024. From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step. arXiv:2405.14838 [cs.CL] https: //arxiv.org/abs/2405.14838

work page internal anchor Pith review Pith/arXiv arXiv 2024
[5]

Yuntian Deng, Kiran Prasad, Roland Fernandez, Paul Smolensky, Vishrav Chaud- hary, and Stuart Shieber. 2023. Implicit Chain of Thought Reasoning via Knowl- edge Distillation. arXiv:2311.01460 [cs.CL] https://arxiv.org/abs/2311.01460

work page arXiv 2023
[6]

Sicheng Feng, Gongfan Fang, Xinyin Ma, and Xinchao Wang. 2025. Efficient Reasoning Models: A Survey.Transactions on Machine Learning Research(2025). https://openreview.net/forum?id=sySqlxj8EB

2025
[7]

Chen Gao, Yu Zheng, Wenjie Wang, Fuli Feng, Xiangnan He, and Yong Li. 2024. Causal Inference in Recommender Systems: A Survey and Future Directions. ACM Trans. Inf. Syst.42, 4, Article 88 (Feb. 2024), 32 pages

2024
[8]

Shijie Geng, Shuchang Liu, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2022. Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5). InProceedings of the 16th ACM Conference on Recommender Systems(Seattle, WA, USA)(RecSys ’22). Association for Computing Machinery, New York, NY, USA, 299–315. doi:10.11...

work page doi:10.1145/3523227.3546767 2022
[9]

Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason E Weston, and Yuandong Tian. 2025. Training Large Language Models to Reason in a Continuous Latent Space. InSecond Conference on Language Modeling. https: //openreview.net/forum?id=Itxz7S4Ip3

2025
[10]

Kalervo Järvelin and Jaana Kekäläinen. 2000. IR evaluation methods for retrieving highly relevant documents. InProceedings of the 23rd Annual International ACM SIGIR Conference on Research and Development in Information Retrieval(Athens, Greece)(SIGIR ’00). Association for Computing Machinery, New York, NY, USA, 41–48. doi:10.1145/345508.345545

work page doi:10.1145/345508.345545 2000
[11]

Wang-Cheng Kang and Julian McAuley. 2018. Self-Attentive Sequential Rec- ommendation . In2018 IEEE International Conference on Data Mining (ICDM). IEEE Computer Society, Los Alamitos, CA, USA, 197–206. doi:10.1109/ICDM. 2018.00035

work page doi:10.1109/icdm 2018
[12]

Chenglin Li, Qianglong Chen, Liangyue Li, Caiyu Wang, Yicheng Li, Zulong Chen, and Yin Zhang. 2024. Mixed Distillation Helps Smaller Language Model Better Reasoning. arXiv:2312.10730 [cs.CL] https://arxiv.org/abs/2312.10730 Generative Spatiotemporal Intent Sequence Recommendation via Implicit Reasoning in Amap

work page arXiv 2024
[13]

Yuetai Li, Xiang Yue, Zhangchen Xu, Fengqing Jiang, Luyao Niu, Bill Yuchen Lin, Bhaskar Ramasubramanian, and Radha Poovendran. 2025. Small Models Struggle to Learn from Strong Reasoners. arXiv:2502.12143 [cs.AI] https://arxiv. org/abs/2502.12143

work page arXiv 2025
[14]

Huanxuan Liao, Shizhu He, Yupu Hao, Xiang Li, Yuanzhe Zhang, Jun Zhao, and Kang Liu. 2024. SKIntern: Internalizing Symbolic Knowledge for Distilling Better CoT Capabilities into Small Language Models. arXiv:2409.13183 [cs.CL] https://arxiv.org/abs/2409.13183

work page arXiv 2024
[15]

Jianghao Lin, Xinyi Dai, Yunjia Xi, Weiwen Liu, Bo Chen, Hao Zhang, Yong Liu, Chuhan Wu, Xiangyang Li, Chenxu Zhu, Huifeng Guo, Yong Yu, Ruiming Tang, and Weinan Zhang. 2025. How Can Recommender Systems Benefit from Large Language Models: A Survey.ACM Trans. Inf. Syst.43, 2, Article 28 (Jan. 2025), 47 pages. doi:10.1145/3678004

work page doi:10.1145/3678004 2025
[16]

Yijia Luo, Yulin Song, Xingyao Zhang, Jiaheng Liu, Weixun Wang, GengRu Chen, Wenbo Su, and Bo Zheng. 2025. Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillation. arXiv:2503.16385 [cs.AI] https://arxiv.org/abs/2503.16385

work page arXiv 2025
[17]

Lucie Charlotte Magister, Jonathan Mallinson, Jakub Adamek, Eric Malmi, and Aliaksei Severyn. 2023. Teaching Small Language Models to Reason. arXiv:2212.08410 [cs.CL] https://arxiv.org/abs/2212.08410

work page arXiv 2023
[18]

Li-Wei Pan, Wei-Ke Pan, Mei-Yan Wei, Hong-Zhi Yin, and Zhong Ming. 2025. A survey on sequential recommendation.Front. Comput. Sci.20, 3 (Oct. 2025), 32 pages. doi:10.1007/s11704-025-41329-w

work page doi:10.1007/s11704-025-41329-w 2025
[19]

Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct Preference Optimization: Your Language Model is Secretly a Reward Model. InThirty-seventh Conference on Neural Infor- mation Processing Systems. https://openreview.net/forum?id=HPuSIXJaa9

2023
[20]

Tran, Jonah Samost, Maciej Kula, Ed H

Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Maheswaran Sathiamoorthy. 2023. Recommender Systems with Generative Retrieval. InThirty-seventh Conference on Neural Information Processing Systems. https://openreview.net/forum?id=B...

2023
[21]

Zhenyi Shen, Hanqi Yan, Linhai Zhang, Zhanghao Hu, Yali Du, and Yulan He
[22]

InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing

CODI: Compressing Chain-of-Thought into Continuous Space via Self- Distillation. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Suzhou, China, 677–693. doi:10.18653/v1/2025.emnlp-main.36

work page doi:10.18653/v1/2025.emnlp-main.36 2025
[23]

Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang
[24]

InProceedings of the 28th ACM International Conference on Information and Knowledge Management(Beijing, China)(CIKM ’19)

BERT4Rec: Sequential Recommendation with Bidirectional Encoder Rep- resentations from Transformer. InProceedings of the 28th ACM International Conference on Information and Knowledge Management(Beijing, China)(CIKM ’19). ACM, New York, NY, USA, 1441–1450. doi:10.1145/3357384.3357895

work page doi:10.1145/3357384.3357895
[25]

Jiakai Tang, Sunhao Dai, Teng Shi, Jun Xu, Xu Chen, Wen Chen, Wu Jian, and Yuning Jiang. 2025. Think Before Recommend: Unleashing the Latent Reasoning Power for Sequential Recommendation. arXiv:2503.22675 [cs.IR] https://arxiv. org/abs/2503.22675

work page arXiv 2025
[26]

Chi, Quoc V

Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2022. Chain-of-thought prompting elicits reasoning in large language models. InProceedings of the 36th International Conference on Neural Information Processing Systems(New Orleans, LA, USA) (NIPS ’22). Curran Associates Inc., Red Hook, NY...

2022
[27]

Likang Wu, Zhi Zheng, Zhaopeng Qiu, Hao Wang, Hongchao Gu, Tingjia Shen, Chuan Qin, Chen Zhu, Hengshu Zhu, Qi Liu, et al . 2024. A survey on large language models for recommendation.World Wide Web27, 5 (2024), 60

2024
[28]

An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jingren Zhou, Junyang Lin, Kai Dang, Keqin Bao, Kexin Yang, ...

work page internal anchor Pith review Pith/arXiv arXiv 2025
[29]

Jiaqi Zhai, Lucy Liao, Xing Liu, Yueming Wang, Rui Li, Xuan Cao, Leon Gao, Zhaojie Gong, Fangda Gu, Michael He, Yinghai Lu, and Yu Shi. 2024. Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Gener- ative Recommendations. InProceedings of the 41st International Conference on Machine Learning (ICML ’24). https://arxiv.org/abs/2...

work page internal anchor Pith review Pith/arXiv arXiv 2024

[1] [1]

Tesfaye Fenta Boka, Zhendong Niu, and Rama Bastola Neupane. 2024. A sur- vey of sequential recommendation systems: Techniques, evaluation, and future directions.Information Systems125 (2024), 102427. doi:10.1016/j.is.2024.102427

work page doi:10.1016/j.is.2024.102427 2024

[2] [2]

Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin...

work page internal anchor Pith review Pith/arXiv arXiv 2020

[3] [3]

Jeffrey Cheng and Benjamin Van Durme. 2024. Compressed Chain of Thought: Efficient Reasoning Through Dense Representations. arXiv:2412.13171 [cs.CL] https://arxiv.org/abs/2412.13171

work page internal anchor Pith review Pith/arXiv arXiv 2024

[4] [4]

Yuntian Deng, Yejin Choi, and Stuart Shieber. 2024. From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step. arXiv:2405.14838 [cs.CL] https: //arxiv.org/abs/2405.14838

work page internal anchor Pith review Pith/arXiv arXiv 2024

[5] [5]

Yuntian Deng, Kiran Prasad, Roland Fernandez, Paul Smolensky, Vishrav Chaud- hary, and Stuart Shieber. 2023. Implicit Chain of Thought Reasoning via Knowl- edge Distillation. arXiv:2311.01460 [cs.CL] https://arxiv.org/abs/2311.01460

work page arXiv 2023

[6] [6]

Sicheng Feng, Gongfan Fang, Xinyin Ma, and Xinchao Wang. 2025. Efficient Reasoning Models: A Survey.Transactions on Machine Learning Research(2025). https://openreview.net/forum?id=sySqlxj8EB

2025

[7] [7]

Chen Gao, Yu Zheng, Wenjie Wang, Fuli Feng, Xiangnan He, and Yong Li. 2024. Causal Inference in Recommender Systems: A Survey and Future Directions. ACM Trans. Inf. Syst.42, 4, Article 88 (Feb. 2024), 32 pages

2024

[8] [8]

Shijie Geng, Shuchang Liu, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2022. Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5). InProceedings of the 16th ACM Conference on Recommender Systems(Seattle, WA, USA)(RecSys ’22). Association for Computing Machinery, New York, NY, USA, 299–315. doi:10.11...

work page doi:10.1145/3523227.3546767 2022

[9] [9]

Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason E Weston, and Yuandong Tian. 2025. Training Large Language Models to Reason in a Continuous Latent Space. InSecond Conference on Language Modeling. https: //openreview.net/forum?id=Itxz7S4Ip3

2025

[10] [10]

Kalervo Järvelin and Jaana Kekäläinen. 2000. IR evaluation methods for retrieving highly relevant documents. InProceedings of the 23rd Annual International ACM SIGIR Conference on Research and Development in Information Retrieval(Athens, Greece)(SIGIR ’00). Association for Computing Machinery, New York, NY, USA, 41–48. doi:10.1145/345508.345545

work page doi:10.1145/345508.345545 2000

[11] [11]

Wang-Cheng Kang and Julian McAuley. 2018. Self-Attentive Sequential Rec- ommendation . In2018 IEEE International Conference on Data Mining (ICDM). IEEE Computer Society, Los Alamitos, CA, USA, 197–206. doi:10.1109/ICDM. 2018.00035

work page doi:10.1109/icdm 2018

[12] [12]

Chenglin Li, Qianglong Chen, Liangyue Li, Caiyu Wang, Yicheng Li, Zulong Chen, and Yin Zhang. 2024. Mixed Distillation Helps Smaller Language Model Better Reasoning. arXiv:2312.10730 [cs.CL] https://arxiv.org/abs/2312.10730 Generative Spatiotemporal Intent Sequence Recommendation via Implicit Reasoning in Amap

work page arXiv 2024

[13] [13]

Yuetai Li, Xiang Yue, Zhangchen Xu, Fengqing Jiang, Luyao Niu, Bill Yuchen Lin, Bhaskar Ramasubramanian, and Radha Poovendran. 2025. Small Models Struggle to Learn from Strong Reasoners. arXiv:2502.12143 [cs.AI] https://arxiv. org/abs/2502.12143

work page arXiv 2025

[14] [14]

Huanxuan Liao, Shizhu He, Yupu Hao, Xiang Li, Yuanzhe Zhang, Jun Zhao, and Kang Liu. 2024. SKIntern: Internalizing Symbolic Knowledge for Distilling Better CoT Capabilities into Small Language Models. arXiv:2409.13183 [cs.CL] https://arxiv.org/abs/2409.13183

work page arXiv 2024

[15] [15]

Jianghao Lin, Xinyi Dai, Yunjia Xi, Weiwen Liu, Bo Chen, Hao Zhang, Yong Liu, Chuhan Wu, Xiangyang Li, Chenxu Zhu, Huifeng Guo, Yong Yu, Ruiming Tang, and Weinan Zhang. 2025. How Can Recommender Systems Benefit from Large Language Models: A Survey.ACM Trans. Inf. Syst.43, 2, Article 28 (Jan. 2025), 47 pages. doi:10.1145/3678004

work page doi:10.1145/3678004 2025

[16] [16]

Yijia Luo, Yulin Song, Xingyao Zhang, Jiaheng Liu, Weixun Wang, GengRu Chen, Wenbo Su, and Bo Zheng. 2025. Deconstructing Long Chain-of-Thought: A Structured Reasoning Optimization Framework for Long CoT Distillation. arXiv:2503.16385 [cs.AI] https://arxiv.org/abs/2503.16385

work page arXiv 2025

[17] [17]

Lucie Charlotte Magister, Jonathan Mallinson, Jakub Adamek, Eric Malmi, and Aliaksei Severyn. 2023. Teaching Small Language Models to Reason. arXiv:2212.08410 [cs.CL] https://arxiv.org/abs/2212.08410

work page arXiv 2023

[18] [18]

Li-Wei Pan, Wei-Ke Pan, Mei-Yan Wei, Hong-Zhi Yin, and Zhong Ming. 2025. A survey on sequential recommendation.Front. Comput. Sci.20, 3 (Oct. 2025), 32 pages. doi:10.1007/s11704-025-41329-w

work page doi:10.1007/s11704-025-41329-w 2025

[19] [19]

Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct Preference Optimization: Your Language Model is Secretly a Reward Model. InThirty-seventh Conference on Neural Infor- mation Processing Systems. https://openreview.net/forum?id=HPuSIXJaa9

2023

[20] [20]

Tran, Jonah Samost, Maciej Kula, Ed H

Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Maheswaran Sathiamoorthy. 2023. Recommender Systems with Generative Retrieval. InThirty-seventh Conference on Neural Information Processing Systems. https://openreview.net/forum?id=B...

2023

[21] [21]

Zhenyi Shen, Hanqi Yan, Linhai Zhang, Zhanghao Hu, Yali Du, and Yulan He

[22] [22]

InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing

CODI: Compressing Chain-of-Thought into Continuous Space via Self- Distillation. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Suzhou, China, 677–693. doi:10.18653/v1/2025.emnlp-main.36

work page doi:10.18653/v1/2025.emnlp-main.36 2025

[23] [23]

Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang

[24] [24]

InProceedings of the 28th ACM International Conference on Information and Knowledge Management(Beijing, China)(CIKM ’19)

BERT4Rec: Sequential Recommendation with Bidirectional Encoder Rep- resentations from Transformer. InProceedings of the 28th ACM International Conference on Information and Knowledge Management(Beijing, China)(CIKM ’19). ACM, New York, NY, USA, 1441–1450. doi:10.1145/3357384.3357895

work page doi:10.1145/3357384.3357895

[25] [25]

Jiakai Tang, Sunhao Dai, Teng Shi, Jun Xu, Xu Chen, Wen Chen, Wu Jian, and Yuning Jiang. 2025. Think Before Recommend: Unleashing the Latent Reasoning Power for Sequential Recommendation. arXiv:2503.22675 [cs.IR] https://arxiv. org/abs/2503.22675

work page arXiv 2025

[26] [26]

Chi, Quoc V

Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2022. Chain-of-thought prompting elicits reasoning in large language models. InProceedings of the 36th International Conference on Neural Information Processing Systems(New Orleans, LA, USA) (NIPS ’22). Curran Associates Inc., Red Hook, NY...

2022

[27] [27]

Likang Wu, Zhi Zheng, Zhaopeng Qiu, Hao Wang, Hongchao Gu, Tingjia Shen, Chuan Qin, Chen Zhu, Hengshu Zhu, Qi Liu, et al . 2024. A survey on large language models for recommendation.World Wide Web27, 5 (2024), 60

2024

[28] [28]

An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jingren Zhou, Junyang Lin, Kai Dang, Keqin Bao, Kexin Yang, ...

work page internal anchor Pith review Pith/arXiv arXiv 2025

[29] [29]

Jiaqi Zhai, Lucy Liao, Xing Liu, Yueming Wang, Rui Li, Xuan Cao, Leon Gao, Zhaojie Gong, Fangda Gu, Michael He, Yinghai Lu, and Yu Shi. 2024. Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Gener- ative Recommendations. InProceedings of the 41st International Conference on Machine Learning (ICML ’24). https://arxiv.org/abs/2...

work page internal anchor Pith review Pith/arXiv arXiv 2024