arxiv: 2605.14512 · v1 · submitted 2026-05-14 · 💻 cs.IR · cs.AI

Recognition: 1 theorem link

· Lean Theorem

Asymmetric Generative Recommendation via Multi-Expert Projection and Multi-Faceted Hierarchical Quantization

Bin Huang , Xin Wang , Junwei Pan , Yongqi Zhou , Yifeng Zhou , Zhixiang Feng , Shudong Huang , Haijie Gu

show 1 more author

Wenwu Zhu

Authors on Pith no claims yet

Pith reviewed 2026-05-15 01:44 UTC · model grok-4.3

classification 💻 cs.IR cs.AI

keywords generative recommendationasymmetric frameworksemantic projectionhierarchical quantizationinformation bottlenecksemantic IDsrecommendation systemstransformer models

0 comments

The pith

An asymmetric continuous-discrete framework removes dual information bottlenecks in generative recommendation and improves accuracy by 15.8 percent on average.

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

Generative recommendation models treat item suggestion as sequence generation but reuse the same discrete Semantic IDs for both model inputs and prediction targets. This creates an input bottleneck from lossy quantization and popularity bias that hurts rare items, plus an output bottleneck from imprecise discrete targets that weaken training signals. The paper introduces AsymRec to break the symmetry by keeping continuous embeddings for inputs and mapping them into the Transformer via multi-expert projections that preserve semantic detail, while constructing richer discrete output targets through multi-view hierarchical quantization with regularization to avoid collapse. Experiments across datasets show consistent gains of 15.8 percent over prior symmetric generative models. A reader would care because the result indicates that recommendation performance can rise by fixing representation mismatch rather than scaling model size.

Core claim

The central claim is that symmetric Semantic ID usage in generative recommenders produces a dual-stage information bottleneck: lossy quantization plus popularity bias degrades input semantics, while imprecise discrete targets limit output supervision. AsymRec decouples the stages with Multi-expert Semantic Projection that routes continuous embeddings through expert-specialized linear maps to retain fine-grained semantics and improve generalization to infrequent items, combined with Multi-faceted Hierarchical Quantization that assembles high-capacity structured targets from multi-view multi-level codes plus semantic regularization to prevent dimensional collapse and keep distinctions intact.

What carries the argument

Multi-expert Semantic Projection (MSP) that maps continuous embeddings into hidden space via expert-specialized projections, paired with Multi-faceted Hierarchical Quantization (MHQ) that builds structured discrete targets through multi-view multi-level encoding and regularization; together they decouple continuous inputs from discrete outputs.

If this is right

Rare items receive better representation because continuous input embeddings avoid quantization loss before projection.
Training signals strengthen because multi-faceted discrete targets supply more precise and structured supervision.
Dimensional collapse is avoided in the quantized space through the combination of multi-view and multi-level quantization plus regularization.
Overall ranking metrics rise consistently, delivering an average 15.8 percent lift over existing generative recommenders.
The same decoupling pattern can be applied without increasing overall model size or inference cost.

Where Pith is reading between the lines

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

The expert-projection idea may transfer to other sequence models that currently force identical discrete tokens for input and output.
Hierarchical quantization with semantic regularization provides a reusable template for building discrete codes in retrieval or ranking tasks beyond recommendation.
Scaling the number of experts or quantization facets on larger catalogs could reveal further gains once the basic asymmetry is in place.
The framework suggests testing whether similar input-output splits help in related generative settings such as session-based prediction.

Load-bearing premise

The identified input and output bottlenecks are the dominant limitations of prior symmetric models and MSP plus MHQ mitigate them without creating new trade-offs in representation quality or training dynamics.

What would settle it

An ablation study on a standard recommendation dataset in which MSP is replaced by a single projection and MHQ by flat quantization produces no gain or a performance drop relative to the symmetric baseline.

Figures

Figures reproduced from arXiv: 2605.14512 by Bin Huang, Haijie Gu, Junwei Pan, Shudong Huang, Wenwu Zhu, Xin Wang, Yifeng Zhou, Yongqi Zhou, Zhixiang Feng.

**Figure 2.** Figure 2: Overview of the proposed AsymRec framework. The input item is first encoded into a continuous semantic embedding, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Retrieval performance at the input stage using Mean [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Normalized Singular Spectrum of Transformer Out [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: NDCG@10 under different quantization configura [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

read the original abstract

Generative Recommendation (GenRec) models reformulate recommendation as a sequence generation task, representing items as discrete Semantic IDs used symmetrically as both inputs and prediction targets. We identify a critical dual-stage information bottleneck in this design: (1) the Input Bottleneck, where lossy quantization degrades fine-grained semantics, while popularity bias skews the learned representations toward frequent items, and (2) the Output Bottleneck, where imprecise discrete targets limit supervision quality. To address these issues, we propose AsymRec, an asymmetric continuous-discrete framework that decouples input and output representations. Specifically, Multi-expert Semantic Projection (MSP) maps continuous embeddings into the Transformer's hidden space via expert-specialized projections, preserving semantic richness and improving generalization to infrequent items. Multi-faceted Hierarchical Quantization (MHQ) constructs high-capacity, structured discrete targets through multi-view and multi-level quantization with semantic regularization, preventing dimensional collapse while retaining fine-grained distinctions. Extensive experiments demonstrate that AsymRec consistently outperforms state-of-the-art generative recommenders by an average of 15.8 %. The code will be released.

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

AsymRec's asymmetric continuous-discrete split with MSP expert projections and MHQ hierarchical quantization is the real novelty here, but the 15.8% average gain is presented without enough experimental detail to confirm it comes from those changes rather than capacity or tuning.

read the letter

The main thing to know is that this paper moves generative recommendation away from symmetric discrete IDs for both input and output. AsymRec keeps continuous embeddings on the input side and maps them through multi-expert projections (MSP) to preserve semantics and help with infrequent items. On the output side it uses multi-faceted hierarchical quantization (MHQ) to build richer discrete targets with regularization against collapse. That split directly targets the input bottleneck from lossy quantization plus popularity skew and the output bottleneck from coarse targets. The motivation lines up with prior symmetric GenRec work, and the specific mechanisms feel like a practical response rather than a vague tweak. Releasing code is a plus for checking the implementation. What the paper does cleanly is name the dual bottlenecks and tie the proposed pieces to them without obvious circularity in the framing. The architecture description is concrete enough that someone could try to replicate the pattern. The soft spot is the evidence. The abstract reports a 15.8% average lift over state-of-the-art generative recommenders, yet supplies no dataset names, baseline list, ablation results, or long-tail breakdowns. Without those, it is difficult to tell whether MSP and MHQ are actually resolving the claimed bottlenecks or whether gains trace to extra parameters, different training, or dataset effects. The stress-test concern holds up on the given material: the headline number alone does not isolate the contribution. If the full paper includes component ablations and per-bottleneck diagnostics, that would change the picture; as presented, the central claim rests on an aggregate that is hard to verify. This is for researchers working on generative models for recommendation who are already dealing with discrete ID limitations. A reader building production systems or running ablations on sequence recommenders would get the most from the architecture details and could test the asymmetry idea themselves. It deserves peer review. The idea is focused and the motivation is honest, so referees can check the experiments, request missing ablations, and decide whether the gains are robust. I would send it forward rather than desk reject.

Referee Report

2 major / 1 minor

Summary. The manuscript identifies dual information bottlenecks in symmetric generative recommendation (GenRec) models—lossy input quantization with popularity skew, and imprecise discrete output targets—and proposes AsymRec, an asymmetric continuous-discrete framework. It introduces Multi-expert Semantic Projection (MSP) to map continuous item embeddings into the Transformer hidden space via specialized experts, and Multi-faceted Hierarchical Quantization (MHQ) to build high-capacity structured discrete targets with multi-view, multi-level quantization and semantic regularization. The central empirical claim is that AsymRec consistently outperforms state-of-the-art generative recommenders by an average of 15.8%.

Significance. If the performance gains are shown to be robust, statistically significant, and directly attributable to the asymmetric design rather than capacity increases, the work would meaningfully advance generative recommendation by offering a concrete mechanism to preserve fine-grained semantics and mitigate popularity bias. The explicit decoupling of input and output representations, together with the planned code release, could serve as a useful baseline for future sequence-based recommenders and encourage further exploration of hybrid continuous-discrete architectures.

major comments (2)

[Experimental results] Experimental results section: The headline claim of an average 15.8% improvement is presented without component ablations, long-tail subset results, or per-bottleneck diagnostics that would isolate whether MSP and MHQ specifically resolve the input/output bottlenecks identified in the introduction rather than other factors such as increased model capacity or training regime differences.
[§3 and §4] §3 (MSP) and §4 (MHQ): The descriptions of expert projections and multi-faceted quantization introduce several free parameters (number of experts, quantization levels and facets) whose sensitivity is not analyzed; without this, it is unclear whether the claimed improvements are stable or require extensive tuning that could undermine the practical advantage over symmetric baselines.

minor comments (1)

[Abstract] Abstract: The statement that AsymRec 'consistently outperforms' SOTA models would be strengthened by briefly indicating the number of datasets and the range of per-dataset gains rather than only the aggregate average.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We agree that strengthening the experimental section with targeted ablations and sensitivity analyses will better substantiate the contributions of the asymmetric design. We outline our responses to each major comment below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [Experimental results] Experimental results section: The headline claim of an average 15.8% improvement is presented without component ablations, long-tail subset results, or per-bottleneck diagnostics that would isolate whether MSP and MHQ specifically resolve the input/output bottlenecks identified in the introduction rather than other factors such as increased model capacity or training regime differences.

Authors: We agree that additional diagnostics are needed to isolate the effects of MSP and MHQ from potential capacity or training differences. In the revised manuscript, we will add component ablations (AsymRec variants with MSP or MHQ removed individually), results on long-tail item subsets, and per-bottleneck metrics such as input reconstruction fidelity and output target precision. These will be presented alongside parameter-matched baselines to confirm that gains arise from the input/output decoupling rather than other factors. revision: yes
Referee: [§3 and §4] §3 (MSP) and §4 (MHQ): The descriptions of expert projections and multi-faceted quantization introduce several free parameters (number of experts, quantization levels and facets) whose sensitivity is not analyzed; without this, it is unclear whether the claimed improvements are stable or require extensive tuning that could undermine the practical advantage over symmetric baselines.

Authors: We acknowledge that sensitivity analysis for the number of experts, quantization levels, and facets would strengthen the practical claims. The revised version will include new experiments varying these hyperparameters across datasets, demonstrating stable performance within practical ranges and that default settings do not require dataset-specific extensive tuning beyond standard validation practices. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain; empirical proposal with external validation

full rationale

The paper identifies input/output bottlenecks in symmetric GenRec conceptually, proposes MSP and MHQ as architectural fixes, and reports aggregate empirical gains (15.8%) against external baselines. No equations, derivations, or self-citations are presented that reduce the claimed improvements to fitted parameters defined by the same data or to prior author results by construction. The result is framed as an empirical comparison, making the central claim self-contained against external benchmarks rather than internally forced.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 2 invented entities

The central claim rests on the domain assumption that symmetric discrete ID usage creates the stated bottlenecks and on two newly introduced architectural modules whose effectiveness is shown empirically rather than derived from first principles.

free parameters (2)

number of experts in MSP
Architectural hyperparameter whose value is chosen to balance capacity and generalization; exact value not stated in abstract.
quantization levels and facets in MHQ
Design choices controlling discrete target capacity and semantic regularization strength.

axioms (1)

domain assumption Symmetric use of discrete Semantic IDs creates both input and output information bottlenecks in generative recommendation.
Invoked in the opening motivation as the critical limitation of prior GenRec designs.

invented entities (2)

Multi-expert Semantic Projection (MSP) no independent evidence
purpose: Maps continuous item embeddings into transformer hidden space via expert-specialized projections to preserve fine-grained semantics.
New component introduced to address the input bottleneck; no independent evidence outside the paper's experiments.
Multi-faceted Hierarchical Quantization (MHQ) no independent evidence
purpose: Constructs high-capacity structured discrete targets via multi-view and multi-level quantization with semantic regularization.
New component introduced to address the output bottleneck; no independent evidence outside the paper's experiments.

pith-pipeline@v0.9.0 · 5517 in / 1395 out tokens · 28585 ms · 2026-05-15T01:44:27.263329+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Multi-expert Semantic Projection (MSP) maps continuous embeddings into the Transformer’s hidden space via expert-specialized projections... Multi-faceted Hierarchical Quantization (MHQ) constructs high-capacity, structured discrete targets through multi-view and multi-level quantization with semantic regularization

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.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · 6 internal anchors

[1]

Prabhat Agarwal, Anirudhan Badrinath, Laksh Bhasin, Jaewon Yang, Edoardo Botta, Jiajing Xu, and Charles Rosenberg. 2025. PinRec: Outcome-Conditioned, Multi-Token Generative Retrieval for Industry-Scale Recommendation Systems. arXiv:2504.10507 [cs.IR] doi:10.48550/arXiv.2504.10507 PinRec

work page doi:10.48550/arxiv.2504.10507 2025
[2]

Gordon V Cormack, Charles LA Clarke, and Stefan Buettcher. 2009. Reciprocal rank fusion outperforms condorcet and individual rank learning methods. In Proceedings of the 32nd international ACM SIGIR conference on Research and development in information retrieval. 758–759

work page 2009
[3]

Jiaxin Deng, Shiyao Wang, Kuo Cai, Lejian Ren, Qigen Hu, Weifeng Ding, Qiang Luo, and Guorui Zhou. 2025. OneRec: Unifying Retrieve and Rank with Genera- tive Recommender and Iterative Preference Alignment. arXiv:2502.18965 [cs.IR] doi:10.48550/arXiv.2502.18965 OneRec

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2502.18965 2025
[4]

Robert Gray. 1984. Vector quantization.IEEE Assp Magazine1, 2 (1984), 4–29

work page 1984
[5]

Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al . 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948(2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025
[6]

Balázs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk

work page
[7]

Session-based recommendations with recurrent neural networks.arXiv preprint arXiv:1511.06939(2015)

work page internal anchor Pith review Pith/arXiv arXiv 2015
[8]

Yupeng Hou, Zhankui He, Julian McAuley, and Wayne Xin Zhao. 2023. Learning vector-quantized item representation for transferable sequential recommenders. InProceedings of the ACM Web Conference 2023. 1162–1171

work page 2023
[9]

Yupeng Hou, Jiacheng Li, Ashley Shin, Jinsung Jeon, Abhishek Santhanam, Wei Shao, Kaveh Hassani, Ning Yao, and Julian McAuley. 2025. Generating long semantic ids in parallel for recommendation. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2. 956–966

work page 2025
[10]

Tianyu Hua, Wenxiao Wang, Zihui Xue, Sucheng Ren, Yue Wang, and Hang Zhao. 2021. On feature decorrelation in self-supervised learning. InProceedings of the IEEE/CVF international conference on computer vision. 9598–9608

work page 2021
[11]

Yanhua Huang, Yuqi Chen, Xiong Cao, Rui Yang, Mingliang Qi, Yinghao Zhu, Qingchang Han, Yaowei Liu, Zhaoyu Liu, Xuefeng Yao, Yuting Jia, Leilei Ma, Yinqi Zhang, Taoyu Zhu, Liujie Zhang, Lei Chen, Weihang Chen, Min Zhu, Ruiwen Xu, and Lei Zhang. 2025. Towards Large-scale Generative Ranking. arXiv:2505.04180 [cs.IR] doi:10.48550/arXiv.2505.04180 GenRank

work page doi:10.48550/arxiv.2505.04180 2025
[12]

Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. 1991. Adaptive mixtures of local experts.Neural computation3, 1 (1991), 79–87

work page 1991
[13]

Michael I Jordan and Robert A Jacobs. 1994. Hierarchical mixtures of experts and the EM algorithm.Neural computation6, 2 (1994), 181–214

work page 1994
[14]

Biing-Hwang Juang and A Gray. 1982. Multiple stage vector quantization for speech coding. InICASSP’82. IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 7. IEEE, 597–600

work page 1982
[15]

Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recom- mendation. In2018 IEEE international conference on data mining (ICDM). IEEE, 197–206

work page 2018
[16]

Doyup Lee, Chiheon Kim, Saehoon Kim, Minsu Cho, and Wook-Shin Han. 2022. Autoregressive Image Generation Using Residual Quantization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 11523–11532

work page 2022
[17]

Xiaopeng Li, Bo Chen, Junda She, Shiteng Cao, You Wang, Qinlin Jia, Haiying He, Zheli Zhou, Zhao Liu, Ji Liu, et al. 2025. A survey of generative recommendation from a tri-decoupled perspective: Tokenization, architecture, and optimization. (2025)

work page 2025
[18]

Chen Ma, Peng Kang, and Xue Liu. 2019. Hierarchical gating networks for sequential recommendation. InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. 825–833

work page 2019
[19]

Julieta Martinez, Holger H Hoos, and James J Little. 2014. Stacked quantizers for compositional vector compression.arXiv preprint arXiv:1411.2173(2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014
[20]

Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton Van Den Hengel

work page
[21]

InProceedings of the 38th international ACM SIGIR conference on research and development in information retrieval

Image-based recommendations on styles and substitutes. InProceedings of the 38th international ACM SIGIR conference on research and development in information retrieval. 43–52

work page
[22]

Petrov and Craig Macdonald

Aleksandr V. Petrov and Craig Macdonald. 2024. RecJPQ: Training Large- Catalogue Sequential Recommenders. InProceedings of the ACM International Conference on Web Search and Data Mining (WSDM). RecJPQ

work page 2024
[23]

Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Tran, Jonah Samost, et al

work page
[24]

Recommender systems with generative retrieval.Advances in Neural Information Processing Systems36 (2023), 10299–10315

work page 2023
[25]

Steffen Rendle, Christoph Freudenthaler, and Lars Schmidt-Thieme. 2010. Fac- torizing personalized markov chains for next-basket recommendation. InPro- ceedings of the 19th international conference on World wide web. 811–820

work page 2010
[26]

Olivier Roy and Martin Vetterli. 2007. The effective rank: A measure of effective dimensionality. In2007 15th European signal processing conference. IEEE, 606–610

work page 2007
[27]

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

work page
[28]

InProceedings of the 28th ACM international conference on information and knowledge management

BERT4Rec: Sequential recommendation with bidirectional encoder rep- resentations from transformer. InProceedings of the 28th ACM international conference on information and knowledge management. 1441–1450

work page
[29]

Jiaxi Tang and Ke Wang. 2018. Personalized top-n sequential recommenda- tion via convolutional sequence embedding. InProceedings of the eleventh ACM international conference on web search and data mining. 565–573

work page 2018
[30]

Aaron van den Oord, Oriol Vinyals, and Koray Kavukcuoglu. 2017. Neural Discrete Representation Learning. InAdvances in Neural Information Processing Systems, Vol. 30. 6306–6315

work page 2017
[31]

Xiaolong Xu, Hongsheng Dong, Lianyong Qi, Xuyun Zhang, Haolong Xiang, Xiaoyu Xia, Yanwei Xu, and Wanchun Dou. 2024. Cmclrec: Cross-modal con- trastive learning for user cold-start sequential recommendation. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1589–1598

work page 2024
[32]

An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025. Qwen3 technical report.arXiv preprint arXiv:2505.09388(2025)

work page internal anchor Pith review Pith/arXiv arXiv 2025
[33]

Jiaqi Zhai, Lucy Liao, Xing Liu, Yueming Wang, Rui Li, Xuan Cao, Leon Gao, Zhaojie Gong, Fangda Gu, Michael He, et al. 2024. Actions speak louder than words: Trillion-parameter sequential transducers for generative recommenda- tions.arXiv preprint arXiv:2402.17152(2024)

work page internal anchor Pith review Pith/arXiv arXiv 2024
[34]

Tingting Zhang, Pengpeng Zhao, Yanchi Liu, Victor S Sheng, Jiajie Xu, Deqing Wang, Guanfeng Liu, Xiaofang Zhou, et al . 2019. Feature-level deeper self- attention network for sequential recommendation.. InIJCAI. 4320–4326

work page 2019
[35]

Guorui Zhou, Hengrui Hu, Hongtao Cheng, Huanjie Wang, Jiaxin Deng, Jinghao Zhang, Kuo Cai, Lejian Ren, Lu Ren, Liao Yu, Pengfei Zheng, Qiang Luo, Qianqian Wang, Qigen Hu, Rui Huang, Ruiming Tang, Shiyao Wang, Shujie Yang, Tao Wu, Wuchao Li, Xinchen Luo, Xingmei Wang, Yi Su, Yunfan Wu, Zexuan Cheng, Zhanyu Liu, Zixing Zhang, Bin Zhang, Boxuan Wang, Chaoyi ...

work page doi:10.48550/arxiv.2508.20900 2025
[36]

Kun Zhou, Hui Wang, Wayne Xin Zhao, Yutao Zhu, Sirui Wang, Fuzheng Zhang, Zhongyuan Wang, and Ji-Rong Wen. 2020. S3-rec: Self-supervised learning for sequential recommendation with mutual information maximization. InPro- ceedings of the 29th ACM international conference on information & knowledge Huang et al. management. 1893–1902

work page 2020
[37]

Yongchun Zhu, Ruobing Xie, Fuzhen Zhuang, Kaikai Ge, Ying Sun, Xu Zhang, Leyu Lin, and Juan Cao. 2021. Learning to warm up cold item embeddings for cold- start recommendation with meta scaling and shifting networks. InProceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1167–1176

work page 2021