Pith. sign in

REVIEW 5 major objections 6 minor 29 references

Feedback-driven profile updates beat open-loop summarization in short-video recommendation.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

RECAP trains a streaming LLM profile updater with GRPO rewards from a dual-tower evaluator, gaining +0.0084 uAUC (cleaned eval) and +0.139% online usage time.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection Solid engineering contribution with a useful design recipe, but the headline closed-loop-vs-open-loop claim is not tested by the online A/B, and the offline evidence is entangled with the reward evaluator's own family. the 5 major comments →

arxiv 2607.15730 v1 pith:D6BDY563 submitted 2026-07-17 cs.IR

RECAP: Feedback-Driven Streaming Semantic User Profiles for Short-Video Recommendation

Ziyi Zhao , Xiaoyou Zhou , Xiao Lv , Yangyang Li , Chubo He , Zhao Liu , Jiayao Shen , Yuqi Liu
show 8 more authors
He Li Chengyi Zhang Jian Liang Ming Li Chongming Gao Fuli Feng Ruiming Tang Han Li
This is my paper
classification cs.IR
keywords user profilinguser modelinglarge language modelsshort video recommendationstreaming profile updateclosed-loop optimizationimplicit feedbackGRPO
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper tries to establish that a user profile for short-video recommendation should be a bounded, structured state updated incrementally and optimized with downstream recommendation feedback, rather than a free-text summary rewritten from scratch. The proposed framework, RECAP, maintains a fixed-size set of interest entries with deterministic lifecycle rules, and trains an LLM-based updater using a reward derived from a dual-tower evaluator trained on label-consistency-filtered behavior pairs. Offline, this closed-loop optimization raises uAUC from 0.7519 to 0.7603 and Recall@2000 from 0.0122 to 0.0128 on cleaned evaluation, and a seven-day online A/B test shows a statistically significant 0.139% gain in average app usage time per user. The paper argues that reward-driven semantic updates generalize better than open-loop summarization because they produce more grounded refinements and user-level abstraction.

Core claim

RECAP's central discovery is that a streaming profile generator can be optimized offline with a recommendation-aligned reward even when the recommender itself is unavailable. Each profile is a bounded structured memory of up to eight interest entries with lifecycle metadata; the LLM emits only a semantic diff (confirmed and new interests), while a deterministic state machine handles strength decay, eviction, and capacity. The reward is built by filtering logged implicit watch/skip pairs with an LLM judge to keep label-consistent examples, training a dual-tower semantic evaluator over rendered profile text and video captions, and using its matching score as a GRPO reward. Applying this closed

What carries the argument

The central mechanism is the decoupling of semantic updates from lifecycle management: the LLM produces only minimal structured semantic diffs, while a deterministic state machine applies update/add/remove operations with strength decay, recency tracking, and capacity control. This bounded structured memory keeps the profile consistent over long streams, and the dual-tower evaluator's matching score, trained on cleaned feedback, serves as the GRPO reward that closes the loop.

Load-bearing premise

The load-bearing premise is that the dual-tower evaluator's matching score, used both as the training reward and as the offline evaluation metric, faithfully reflects how useful a profile is for real recommendation; if optimizing it mainly aligns profile embeddings with captions rather than improving user satisfaction, the offline gains could be artifacts.

What would settle it

Run a deployment A/B test in which the profile updater is trained on a reward from a freshly trained evaluator using post-deployment interaction logs, and check whether the offline cleaned-evaluation gains persist. If they shrink or invert, the original evaluator reward was overfit. Alternatively, compare RECAP against the base generator on ranking metrics produced by the production recommender itself, rather than by the dual-tower evaluator.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Cleaned label-consistent feedback provides a more stable reward than raw implicit labels, leading to more grounded profile edits and fewer empty or unsupported interest entries.
  • SFT initialization and GRPO optimization work complementarily: SFT improves behavioral coverage and interest separation, while GRPO shapes how the profile is updated and abstracted.
  • The decoupled design (LLM semantics plus deterministic state machine) avoids lifecycle-field errors that a coupled full-state updater introduces.
  • Larger profile capacity improves offline profile utility but at higher generation and encoding cost; eight interest entries balance diversity and compactness.
  • Feeding RECAP-rendered profiles as text embeddings to a production retrieval model yields a statistically significant 0.139% increase in average usage time per user.

Where Pith is reading between the lines

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

  • If the dual-tower evaluator is a faithful reward proxy, the same closed-loop recipe could transfer to other recommendation surfaces where items have semantic captions and implicit feedback is logged.
  • A testable extension: replacing the LLM-judge pairwise filtering with a learned consistency model could reduce dependence on a large judge at scale and enlarge the usable feedback set.
  • The bounded structured profile is a reusable user state for tasks beyond retrieval, such as ranking, explanation, or cross-domain transfer, since it is an interpretable text artifact.
  • Because the reward is a proxy trained on logged behavior, gains may plateau once profile text stops changing the evaluator's score; periodically retraining the evaluator on new interactions could sustain improvements.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes RECAP, an offline closed-loop framework for optimizing streaming structured semantic user profiles in short-video recommendation. Profiles are maintained as bounded slot-based interest entries, with an LLM producing semantic diffs and a deterministic state machine handling lifecycle and capacity control. To obtain a recommendation-aligned reward, the paper filters implicit behavior pairs with an LLM judge, trains a dual-tower evaluator over rendered profiles and video captions, and uses the evaluator's matching score as a GRPO reward for policy optimization. Offline experiments on internal Kuaishou data report cleaned-eval uAUC gains from 0.7519 to 0.7603 and Recall@2000 from 0.0122 to 0.0128 for RECAP over Base, with smaller raw-eval gains, and a seven-day online A/B test reports a statistically significant +0.139% relative improvement in average application usage time per user.

Significance. If the closed-loop optimization result is substantiated, the paper would make a useful contribution to LLM-based user profiling: it treats the profile as a stateful bounded memory, separates semantic updates from deterministic bookkeeping, and demonstrates that feedback construction quality matters for GRPO-based profile optimization. The paper has real strengths: a concrete industrial setting, a held-out user set for offline evaluation, a disjoint evaluator-instance protocol, a controlled comparison of raw vs. cleaned feedback rewards, and rich profile-level diagnostics that connect reward signal quality to update behavior. These go beyond many purely offline LLM-profile papers. The main weakness is that the central claim — closed-loop optimization over an open-loop generator — is supported primarily by an offline evaluator that shares its dual-tower architecture and label-cleaning procedure with the training reward, while the online A/B test does not include an open-loop profile arm. With an additional control arm or a reframed claim, the contribution could be solid, but the current evidence is insufficient for the headline generalization claim.

major comments (5)
  1. [§5.5, Table 2] The online A/B test compares RECAP against a no-profile control, not against Base or SFT. The reported +0.139% usage-time gain therefore supports 'adding a semantic profile feature to retrieval helps,' not 'GRPO closed-loop optimization beats open-loop summarization.' The only direct RECAP-vs-Base comparison is Table 2, and that comparison is entangled with the training reward (see next comment). To support the central claim, the paper should either add an A/B arm serving the Base or SFT profile, or explicitly restrict the online claim to 'profile feature helps' and present RECAP-vs-Base as an offline-only result.
  2. [§4.3.3 Eq. (5), §5.1] The GRPO reward is the negative BCE loss plus cosine margin of a dual-tower evaluator over rendered profile text and video captions. The final offline evaluation uses the same dual-tower architecture and the same LLM-judge label-consistency cleaning to construct training and evaluation pairs. Because the policy is optimized against the same scoring family used for evaluation, the cleaned-eval gains (0.7519→0.7603 uAUC) may partly reflect better alignment to that scoring family rather than better recommendation utility. The held-out user split and separately trained evaluators are good, but they do not break the architectural and label-construction correlation. An independent evaluation signal — e.g., a production ranker score, a different architecture, or a downstream online outcome measured per profile variant — is needed for the RECAP-vs-Base claim.
  3. [§5.2, Table 2] No error bars, confidence intervals, or significance tests are reported for the offline metrics. The raw-evaluation improvements are very small: uAUC 0.5871→0.5886 and Recall 0.0088→0.0089, while GRPO-raw decreases Recall relative to Base. Without variance estimates across seeds or user subsets, it is impossible to tell whether the raw-eval differences are noise. Since the abstract emphasizes the 0.0084 cleaned-uAUC improvement, the paper should also make explicit that this is on cleaned evaluation and should report whether the raw-eval differences are statistically distinguishable.
  4. [§4.4, Eq. (7)–(10)] The parallel chunk-level rollout trains the policy only on diffs conditional on SFT-computed prefix profiles, while during inference the optimized policy's own outputs are fed forward sequentially through the state machine. This train/serve distribution shift is never analyzed. The approximation is justified by cost, but the paper should quantify its effect, e.g., by comparing against sequential GRPO on a subset of users or by measuring prefix-profile drift. Without this, the streaming objective in Eq. (1) is only loosely connected to the actual training objective.
  5. [§4.3.2, §5.1] The label-consistency threshold τ is set to 0.45, which is barely above chance for a binary A/B judge. No analysis is provided of judge agreement, threshold sensitivity, or the resulting cleaning rate. Because the cleaned-feedback reward is a core contribution, the paper should report how performance varies with τ and ideally evaluate the LLM judge's accuracy against a small labeled sample. Relatedly, the penalty weights λ_fmt and λ_div in Eq. (8) are never reported; they should be listed in the implementation details.
minor comments (6)
  1. [§5.1] The values of λ_fmt, λ_div, and ε_clip are missing from the implementation details. The paper states w_rec=1 and w_bce=1, w_nce=0.1, but not the Eq. (8) penalty weights.
  2. [§5.2, Figure 2] Bar plots and line plots in Figure 2 do not include error bars or any variance information. For a paper whose headline offline differences are small, this is important for interpretation.
  3. [§5.5] The A/B test is described only as 'statistically significant' with +0.139%. Report the confidence interval, p-value, or the test used, and state the number of users or traffic volume in each arm beyond the 10%/20% allocation.
  4. [Figure 4] The bar labels 'Raw', 'Clean', and 'S+G' are used in the caption but not defined there; the caption should state that these abbreviate GRPO-raw, GRPO-clean, and SFT+GRPO.
  5. [§4.3.2] The description of the LLM judge says 'the first valid A/B token is parsed as the pairwise preference,' but no prompt template or parsing details are given. Since the cleaning pipeline is central, an appendix with the prompt and a few judged examples would help reproducibility.
  6. [§5.4] The diagnostics use 'gpt-oss-120b' as an LLM judge for coverage and semantic annotation, but no calibration or agreement with human annotation is reported. A small human-annotated validation set would strengthen these analyses.

Circularity Check

1 steps flagged

Offline evaluation is an in-family test of the same dual-tower score used as GRPO reward; online A/B lacks the Base control.

specific steps
  1. fitted input called prediction [Sec. 4.3.3 Eq. (5); Sec. 5.1 Training and evaluation protocol]
    "The recommendation-aligned reward is defined as R_rec(˜P^t_{u,i}) = exp(−L_bce(s(˜P^t_{u,i}), V_u)) + w_rec·Δcos(...) (Eq. 5); "We use the same dual-tower architecture for reward construction and final offline evaluation, but instantiate and train separate evaluators on disjoint user sets." (Sec. 5.1)."

    The GRPO policy is optimized to maximize R_rec, which is a dual-tower evaluator's matching score. The offline 'profile utility' metrics (uAUC, Recall@2000) are computed by another dual-tower evaluator with the same architecture and the same cleaned/raw data definitions. Thus the headline offline gain (cleaned uAUC 0.7519→0.7603) is an in-family evaluation of the same scoring function family used as the training reward. Held-out users and separate evaluator instances make this a generalization test within that family, not an independent test of real recommendation utility; raw-eval gains are much smaller (uAUC 0.5871→0.5886), and the online A/B lacks a Base-profile control arm, so the central closed-loop-vs-open-loop claim rests on this entangled offline evaluation.

full rationale

The main circularity is the dual-tower evaluator doubling as both the GRPO reward (Eq. 5) and the offline evaluation metric (Sec. 5.1). The paper is transparent about using the same architecture but with separate evaluators on disjoint user sets; this reduces but does not remove the construction-level overlap. The policy is directly optimized to maximize a combination of BCE and cosine-margin scores from the evaluator family, so observing improved uAUC/Recall on a freshly trained instance of the same family on similarly filtered data is partly a self-fulfilling outcome. The raw-eval numbers are markedly smaller, and the seven-day online A/B compares RECAP (profile feature) to a no-profile control, not to the Base generator, so it does not independently validate the central closed-loop-vs-open-loop claim. I found no load-bearing self-citation, no imported uniqueness theorem, and no ansatz smuggled via citation; the remaining weaknesses are experimental-design gaps rather than additional circularity. Score 5 reflects one central prediction that is partially forced by construction while still having some held-out generalization content.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 1 invented entities

The central claim rests on several unverified modeling choices: the LLM judge's q_cons filtering, the dual-tower evaluator as reward and metric, and the SFT-prefix approximation of streaming rollouts. Free parameters are mostly hand-set without sensitivity analysis; notably τ=0.45, L=8, and unreported λ_fmt/λ_div. No new physical entities are introduced; q_cons is an in-paper construct without external validation.

free parameters (6)
  • label-consistency threshold τ = 0.45
    Samples with q_cons > τ are retained as cleaned behavioral data; threshold is hand-set and no sensitivity analysis is shown (Sec. 5.1).
  • profile capacity L = 8
    Chosen by ablation balancing performance and cost; all main results use L=8 (Sec. 5.3.2).
  • cosine-margin reward weight w_rec = 1
    Weights the Δ_cos term in Eq. (5); set to 1 in all experiments without a sensitivity analysis.
  • evaluator loss weights w_bce, w_nce = 1, 0.1
    Weights in Eq. (4); fixed by choice, no sensitivity analysis.
  • format/divergence penalty weights λ_fmt, λ_div = not reported
    Appear in Eq. (8) but their values are never given; they directly shape the optimized objective.
  • chunk size B / history length N = 50 / 500
    Production-motivated design choices that define the streaming setup; not varied (Sec. 3.2).
axioms (5)
  • domain assumption LLM judge preferences are a valid measure of label consistency.
    RECAP's cleaned feedback and rewards depend on q_cons from Qwen3-30B pairwise judgments (Sec. 4.3.2); no validation that judge agreement corresponds to true semantic preference.
  • domain assumption Dual-tower evaluator score is a valid proxy for recommendation utility.
    Eq. (5) uses evaluator matching as the GRPO reward and Sec. 5.1 uses the same family as offline metric; the paper never validates the proxy against production ranking except the one online metric.
  • domain assumption SFT-computed prefix profiles accurately stand in for online streaming state.
    Parallel chunk-level rollouts use SFT-generated previous states Pbar_{t-1}; this assumes the SFT state distribution matches what the learned policy would produce (Sec. 4.4).
  • domain assumption Effective view (play >7s or completion) is a valid positive label; exposed-but-not-effective is a valid negative.
    Task definition in Sec. 3.2 inherits production engagement criterion; implicit labels may conflate content form, context, and inertia, which is why cleaning is needed.
  • domain assumption Generated captions (tag, desc) faithfully represent video content.
    Captions from Qwen3-VL-8B are the input to both profile updates and reward/evaluation (Sec. 3.1); caption errors propagate.
invented entities (1)
  • label-consistency score q_cons no independent evidence
    purpose: Filter logged positive/negative behavior pairs into high-confidence cleaned feedback used to train the evaluator and compute rewards.
    q_cons is defined by the paper's LLM judge with a hand-set τ=0.45 and is not validated against external labels; it is an in-paper construct with no falsifiable handle outside the method.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of RECAP: Feedback-Driven Streaming Semantic User Profiles for Short-Video Recommendation." pith.science (2026). https://pith.science/paper/D6BDY563

@misc{pith2026260715730,
  author       = {Pith},
  title        = {Pith review of: RECAP: Feedback-Driven Streaming Semantic User Profiles for Short-Video Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D6BDY563}},
  note         = {Machine review of arXiv:2607.15730}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Language-based user profiles convert long behavioral histories into explicit semantic representations for recommendation. However, most profile generators are optimized in an open loop: they may summarize past behavior fluently, but are not directly trained to improve future recommendation. We study this problem in real-world short-video recommendation, where user behaviors continuously arrive as streams and profiles must be incrementally updated under limited capacity. This requires maintaining a consistent bounded profile state and constructing profile-targeted semantic feedback from industrial implicit behavior logs. We propose RECAP, an offline closed-loop framework for optimizing streaming structured semantic profiles with historical recommendation feedback. RECAP maintains each profile as a bounded structured memory by combining LLM-based semantic updates with deterministic lifecycle and capacity control. RECAP constructs profile-targeted semantic feedback by filtering label-consistent behavior pairs with an LLM judge and training a dual-tower evaluator whose matching score serves as a GRPO reward. Experiments on Kuaishou short-video data show that RECAP improves uAUC by 0.0084 and Recall@2000 by about 4.9% over the base generator. Further analyses confirm the benefits of feedback construction and policy optimization, and show more grounded refinement and user-level abstraction in profile updates. A seven-day online A/B test further shows a statistically significant 0.139% improvement in average application usage time per user.

Figures

Figures reproduced from arXiv: 2607.15730 by Chengyi Zhang, Chongming Gao, Chubo He, Fuli Feng, Han Li, He Li, Jian Liang, Jiayao Shen, Ming Li, Ruiming Tang, Xiao Lv, Xiaoyou Zhou, Yangyang Li, Yuqi Liu, Zhao Liu, Ziyi Zhao.

Figure 1
Figure 1. Figure 1: Overview of the RECAP framework. machine control redundancy, but are not rendered into the pro￾file. A new entry is inserted only when supported by at least two videos, because insertion consumes limited profile capacity and thus requires stronger evidence than updating an existing entry. State machine SM(·) (no learnable parameters). The state ma￾chine applies three operations to the LLM output and 𝑃 𝑡−1 … view at source ↗
Figure 2
Figure 2. Figure 2: Ablation studies of RECAP. (a) compares different reward formulations. (b) evaluates different profile capacities [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison of generated profile descrip [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Profile behavior diagnostics under different training signals. (a) Coverage is the LLM-judged fraction of videos in each [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: LLM-as-judge annotation of GRPO-clean and GRPO [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

29 extracted references · 1 canonical work pages

  1. [1]

    Zhaolin Gao, Joyce Zhou, Yijia Dai, and Thorsten Joachims. 2025. End- to-end Training for Recommendation with Language-based User Profiles. arXiv:2410.18870 [cs.IR] https://arxiv.org/abs/2410.18870

    Pith/arXiv arXiv 2025

  2. [2]

    Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural Collaborative Filtering. InProceedings of the 26th International Conference on World Wide Web(Perth, Australia)(WWW ’17). International World Wide Web Conferences Steering Committee, Republic and Canton of Geneva, CHE, 173–182. doi:10.1145/3038912.3052569

    arXiv 2017

  3. [3]

    Yifan Hu, Yehuda Koren, and Chris Volinsky. 2008. Collaborative Filtering for Implicit Feedback Datasets. In2008 Eighth IEEE International Conference on Data Mining. 263–272. doi:10.1109/ICDM.2008.22

    doi:10.1109/icdm.2008.22 2008

  4. [4]

    Jiazheng Kang, Mingming Ji, Zhe Zhao, and Ting Bai. 2025. Memory OS of AI Agent. InProceedings of the 2025 Conference on Empirical Methods in Natural Lan- guage Processing, Christos Christodoulopoulos, Tanmoy Chakraborty, Carolyn Rose, and Violet Peng (Eds.). Association for Computational Linguistics, Suzhou, China, 25961–25970. doi:10.18653/v1/2025.emnlp...

    doi:10.18653/v1/2025.emnlp-main.1318 2025

  5. [5]

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

    arXiv 2018

  6. [7]

    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

    doi:10.1145/3678004 2025

  7. [8]

    OpenAI. 2025. gpt-oss-120b & gpt-oss-20b Model Card. arXiv:2508.10925 [cs.CL] https://arxiv.org/abs/2508.10925

    Pith/arXiv arXiv 2025

  8. [9]

    Bernstein

    Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2023. Generative Agents: Interactive Simulacra of Human Behavior. InProceedings of the 36th Annual ACM Symposium on User Interface Software and Technology(San Francisco, CA, USA)(UIST ’23). Association for Computing Machinery, New York, NY, USA, ...

    arXiv 2023

  9. [10]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision. InProceedings of the 38th Inter- national Conference on Machine Learning (Proceedings of Machi...

    2021

  10. [11]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y. K. Li, Y. Wu, and Daya Guo. 2024. DeepSeek- Math: Pushing the Limits of Mathematical Reasoning in Open Language Models. arXiv:2402.03300 [cs.CL] https://arxiv.org/abs/2402.03300

    Pith/arXiv arXiv 2024

  11. [12]

    Yunxiao Shi, Wujiang Xu, Zhang Zeqi, Xing Zi, Qiang Wu, and Min Xu. 2025. PersonaX: A Recommendation Agent-Oriented User Modeling Framework for Long Behavior Sequence. InFindings of the Association for Computational Linguis- tics: ACL 2025, Wanxiang Che, Joyce Nabende, Ekaterina Shutova, and Moham- mad Taher Pilehvar (Eds.). Association for Computational ...

    work page doi:10.18653/v1/2025.findings-acl.300 2025

  12. [13]

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

  13. [14]

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. 2019. Representation Learning with Contrastive Predictive Coding. arXiv:1807.03748 [cs.LG] https://arxiv.org/ abs/1807.03748

    Pith/arXiv arXiv 2019

  14. [15]

    Lu Wang, Di Zhang, Fangkai Yang, Pu Zhao, Jianfeng Liu, Yuefeng Zhan, Hao Sun, Qingwei Lin, Weiwei Deng, Dongmei Zhang, Feng Sun, and Qi Zhang

  15. [16]

    Sheng, and Mehmet Orgun

    Shoujin Wang, Liang Hu, Yan Wang, Longbing Cao, Quan Z. Sheng, and Mehmet Orgun. 2019. Sequential Recommender Systems: Challenges, Progress and Prospects. InProceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI-19. International Joint Conferences on Artificial Intelligence Organization, 6332–6338. doi:10.24963/i...

    doi:10.24963/ijcai.2019/883 2019

  16. [17]

    Yunjia Xi, Weiwen Liu, Jianghao Lin, Xiaoling Cai, Hong Zhu, Jieming Zhu, Bo Chen, Ruiming Tang, Weinan Zhang, and Yong Yu. 2024. Towards Open-World Recommendation with Knowledge Augmentation from Large Language Models. InProceedings of the 18th ACM Conference on Recommender Systems(Bari, Italy) (RecSys ’24). Association for Computing Machinery, New York,...

    arXiv 2024

  17. [18]

    Yu Xia, Rui Zhong, Hao Gu, Wei Yang, Chi Lu, Peng Jiang, and Kun Gai. 2025. Hierarchical Tree Search-based User Lifelong Behavior Modeling on Large Language Model. InProceedings of the 48th International ACM SIGIR Confer- ence on Research and Development in Information Retrieval(Padua, Italy)(SIGIR ’25). Association for Computing Machinery, New York, NY, ...

    arXiv 2025

  18. [19]

    Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, and Yongfeng Zhang

  19. [20]

    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, ...

    Pith/arXiv arXiv 2025

  20. [21]

    Chao Yi, Dian Chen, Gaoyang Guo, Jiakai Tang, Jian Wu, Jing Yu, Mao Zhang, Wen Chen, Wenjun Yang, Yujie Luo, Yuning Jiang, Zhujin Gao, Bo Zheng, Binbin Cao, Changfa Wu, Dixuan Wang, Han Wu, Haoyi Hu, Kewei Zhu, Lang Tian, Lin Yang, Qiqi Huang, Siqi Yang, Wenbo Su, Xiaoxiao He, Xin Tong, Xu Chen, Xunke Xi, Xiaowei Huang, Yaxuan Wu, Yeqiu Yang, Yi Hu, Yujin...

    arXiv 2025

  21. [22]

    arXiv:2502.12110 [cs.CL] https://arxiv.org/abs/2502.12110

    A-MEM: Agentic Memory for LLM Agents. arXiv:2502.12110 [cs.CL] https://arxiv.org/abs/2502.12110

    Pith/arXiv arXiv

  22. [23]

    Yongfeng Zhang and Xu Chen. 2020. Explainable Recommendation: A Survey and New Perspectives.Found. Trends Inf. Retr.14, 1 (March 2020), 1–101. doi:10. 1561/1500000066

    2020

  23. [24]

    Zihuai Zhao, Wenqi Fan, Jiatong Li, Yunqing Liu, Xiaowei Mei, Yiqi Wang, Zhen Wen, Fei Wang, Xiangyu Zhao, Jiliang Tang, and Qing Li. 2024. Recommender Systems in the Era of Large Language Models (LLMs).IEEE Trans. on Knowl. and Data Eng.36, 11 (Nov. 2024), 6889–6907. doi:10.1109/TKDE.2024.3392335

    arXiv 2024

  24. [25]

    Chao Yi, Dian Chen, Gaoyang Guo, Jiakai Tang, Jian Wu, Jing Yu, Mao Zhang, Sunhao Dai, Wen Chen, Wenjun Yang, Yuning Jiang, Zhujin Gao, Bo Zheng, Chi Li, Dimin Wang, Dixuan Wang, Fan Li, Fan Zhang, Haibin Chen, Haozhuang Liu, Jialin Zhu, Jiamang Wang, Jiawei Wu, Jin Cui, Ju Huang, Kai Zhang, Kan Liu, Lang Tian, Liang Rao, Longbin Li, Lulu Zhao, Na He, Pei...

    Pith/arXiv arXiv 2025

  25. [26]

    Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, and Yanlin Wang. 2024. MemoryBank: enhancing large language models with long-term memory. In Proceedings of the Thirty-Eighth AAAI Conference on Artificial Intelligence and Thirty-Sixth Conference on Innovative Applications of Artificial Intelligence and Fourteenth Symposium on Educational Advances in Artifici...

    doi:10.1609/aaai 2024

  26. [27]

    Chenxu Zhu, Shigang Quan, Bo Chen, Jianghao Lin, Xiaoling Cai, Hong Zhu, Xiangyang Li, Yunjia Xi, Weinan Zhang, and Ruiming Tang. 2024. LIBER: Lifelong User Behavior Modeling Based on Large Language Models. arXiv:2411.14713 [cs.IR] https://arxiv.org/abs/2411.14713

    Pith/arXiv arXiv 2024

  27. [28]

    Zhi Zheng, WenShuo Chao, Zhaopeng Qiu, Hengshu Zhu, and Hui Xiong. 2024. Harnessing Large Language Models for Text-Rich Sequential Recommendation. InProceedings of the ACM Web Conference 2024(Singapore, Singapore)(WWW ’24). Association for Computing Machinery, New York, NY, USA, 3207–3216. doi:10.1145/3589334.3645358

    arXiv 2024

  28. [2019]

    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

    arXiv

  29. [2025]

    InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2(Toronto ON, Canada)(KDD ’25)

    LettinGo: Explore User Profile Generation for Recommendation System. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2(Toronto ON, Canada)(KDD ’25). Association for Computing Machinery, New York, NY, USA, 2985–2995. doi:10.1145/3711896.3737024

    arXiv

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.