Pith. sign in

REVIEW 4 major objections 5 minor 50 references

RecGPT: A Foundation Model for Sequential Recommendation

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read RecGPT claims that a transformer pre-trained on item text alone can recommend in unseen domains with no target-domain training, beating ID-based recommenders that saw 10% of the data.

desk verdict A genuinely new text-driven tokenization architecture and the right zero-shot evaluation design, undercut by an abstract that overclaims consistency and a scaling-law section that fits instead of predicts. read the letter →

arxiv 2506.06270 v2 pith:RMF6RQRP submitted 2025-06-06 cs.IR

Yangqin Jiang , Xubin Ren , Lianghao Xia , Da Luo , Kangyi Lin , Chao Huang This is my paper
classification cs.IR
keywords sequentialrecommendationzero-shotgeneralizationfoundationmodelFiniteScalarQuantizationtext-drivenitemrepresentationcross-domainautoregressivecold-start
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

RecGPT tries to establish that a recommender system can generalize to unseen domains without any target-domain training, by replacing item IDs with tokens derived solely from item text. The paper argues that pre-training an autoregressive transformer on interaction sequences from eleven e-commerce categories yields true zero-shot ability: on six held-out datasets, RecGPT outperforms recommenders that learn separate embeddings for each item ID and were trained on 10% of the target-domain data, and the same model also works on an industrial news platform. If the claim holds, cold-start and cross-platform recommendation would no longer require per-domain retraining, because any new item can be embedded from its description at inference time. The paper also reports that the model's loss follows an LLM-style power-law scaling curve, with a sharp improvement in zero-shot performance between 10% and 25% of pre-training data.

What carries the argument

The load-bearing machinery is a text-only item tokenization pipeline: MPNet maps each item's title and category text to a continuous embedding; Finite Scalar Quantization (FSQ), a rounding-based scheme, splits that embedding into four sub-vectors and rounds them into discrete tokens from a shared codebook of 15,360 tokens, so any item with text can be tokenized without retraining. On top of this, a transformer with hybrid attention lets tokens within the same item attend bidirectionally while keeping causal attention across items, preserving both item coherence and sequence order. An auxiliary stream adds the continuous semantic embeddings, layer-normalized and added to the discrete token embeddings plus positional embeddings, to compensate for quantization loss. At inference, a catalog-aware beam search with Trie-based prefix constraints searches only token sequences that correspond to real catalog items, predicting all K tokens of the next item in one pass.

What would settle it

Permute item text fields randomly while keeping user interaction sequences identical on a held-out dataset such as Yelp or Steam. If RecGPT's zero-shot Hit@5 stays near its reported level, the model is not actually using text semantics; if it collapses toward the random-token ablation (w/o FSQ), the claim that text drives transfer is confirmed. A complementary test is to pre-train on the same eleven categories with descriptions replaced by random strings and check whether zero-shot performance on Baby, Yelp, and Steam drops to chance.

Watch

Extended reading notes

Core claim

The paper's central claim is that text-driven, token-based autoregressive modeling is sufficient for cross-domain sequential recommendation. RecGPT represents each item by quantizing MPNet text embeddings into a small number of discrete tokens via Finite Scalar Quantization, treats a user's history as a token sequence, and trains on next-token prediction. Without any fine-tuning, it outperforms GRU4Rec, Caser, BERT4Rec, FDSA, CL4SRec, DuoRec, ICLRec, and MAERec trained on 10% of the target data across Baby, Games, Office, Yelp, Washington, and Steam, and also beats S3-Rec, UniSRec, VQ-Rec, TIGER, IDGenRec, and RecFormer in zero-shot comparisons. The authors further claim that ablation results isolate the semantic tokenizer, the hybrid bidirectional-causal attention, and the auxiliary continuous-semantic stream as individually necessary, and that a controlled comparison with VQ-Rec attributes the advantage to architecture rather than data scale. In addition, RecGPT shows power-law scaling of evaluation loss with token count and an emergent generalization threshold, both taken as evidence that recommendation foundation models obey scaling laws analogous to language models.

Load-bearing premise

The load-bearing premise is that textual descriptions are sufficient, domain-invariant signals of user preference; if title-and-category text does not encode the behavioral or platform-specific signals behind choices, the zero-shot mechanism collapses.

Editorial extensions

If this is right

  • New items with a textual description can enter the catalog at inference time, making item cold-start immediate and removing the need for per-domain ID tables.
  • A single pre-trained model can serve e-commerce, local-business, gaming, and news recommendation without fine-tuning, at least when text carries the preference signal.
  • Pre-training data volume, not model size, is the cheaper lever: the reported power-law scaling means performance can be forecast and improved by adding tokens.
  • The beam-search decoder keeps inference fast enough for production by pruning the token space with the catalog trie.
  • The hybrid attention design is necessary: ablations show removing bidirectional intra-item attention or the auxiliary semantic stream degrades zero-shot performance.

Reading between the lines

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

  • A direct extension of the tokenization logic: swapping the text encoder for a multimodal encoder should make the same zero-shot recipe work for music, images, and video items, which the paper lists as future work.
  • A boundary condition follows: on domains where choices are driven by visual or audio attributes, price, or platform-specific popularity rather than title and category text, the zero-shot advantage should shrink as the text-to-behavior correlation falls.
  • The scaling-law result is extrapolatable: training on more categories or languages should lower the 10%-to-25% threshold where zero-shot ability jumps, a prediction that could be tested by measuring the discontinuity under different pre-training mixtures.
  • Because the model is text-only, user cold-start still lacks profile signals such as age and location; combining text tokens with lightweight user features is a natural hybrid that preserves transfer while adding the signals the paper concedes are missing.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. This paper proposes RecGPT, a decoder-only transformer for sequential recommendation that replaces item IDs with text-derived discrete tokens. Item text is embedded with MPNet, quantized with Finite Scalar Quantization into K codebook tokens, and user histories are modeled with hybrid bidirectional (intra-item) and causal (inter-item) attention plus an auxiliary continuous embedding stream. RecGPT is pre-trained on eleven Amazon categories and evaluated zero-shot on three held-out Amazon categories and three non-Amazon platforms (Yelp, Washington, Steam), with additional cold-start, ablation, scaling-law, and industrial-dataset experiments. The core claim is that without any target-domain training, RecGPT significantly and consistently outperforms ID-based baselines trained on 10% of target-domain data.

Significance. The evaluation design has real strengths: the main zero-shot test sets are external to the pre-training corpus, so the central generalization claim is not circular; source code is released; and the scaling-law analysis is a falsifiable, practically useful target. If the comparative claims were supported, the text-only tokenization and catalog-aware beam search would be a noteworthy step toward cross-domain and cold-start recommendation. However, the manuscript's own tables contradict the 'consistent performance advantages' claim on several Hit@5 cells, and the absence of repeated-run statistics makes the significance assertions unverifiable.

major comments (4)
  1. [3.2, Table 1; Abstract] The abstract and Section 3.2 state that RecGPT 'significantly outperforms' and shows 'consistent performance advantages' over 10%-trained baselines, but Table 1 contains direct counterexamples. On Steam, FDSA reaches Hit@5 0.1366 versus RecGPT's 0.1253; on Washington, MAERec reaches Hit@5 0.0154 versus RecGPT's 0.0130. In Table 2's cold-start setting, RecGPT's Hit@5 is below DuoRec on Baby (0.0172 vs 0.0204), below CL4SRec on Yelp (0.0131 vs 0.0221), and below DuoRec on Office (0.0204 vs 0.0207). These cells should either be acknowledged as exceptions or the claim should be revised to a per-metric statement such as 'generally higher NDCG@5 with mixed Hit@5 outcomes.' The wording 'consistent' and 'significantly outperforming' is not supported by the evidence shown.
  2. [3.2, Tables 1-2] No standard deviations, confidence intervals, or repeated-seed results are reported anywhere in the tables, despite the footnote claiming p<0.05 significance marked by '*'. In the absence of error bars, the stars appear to denote only that the RecGPT value is higher than the best baseline in that row, which is not a significance test. Please report mean and standard deviation over at least three seeds and specify the statistical test used, or remove the significance claim.
  3. [3.6, Fig. 5] The scaling-law section fits a power-law curve to evaluation losses of the 5%, 10%, 25%, 50%, and 100% variants and states that it can 'accurately predict the performance of the full model,' but no equation, fitted exponent, goodness of fit, or prediction error is given. It is also unclear whether the solid curve in Figure 5 is the fitted power law or an interpolation between points. Please report the fitted functional form, its parameters, R^2 or residual error, and the predicted-versus-observed value for the full model; otherwise the power-law claim is not checkable.
  4. [3.5, Table 3] The 'w/o Pre' ablation removes the Trie-based prefix enhancement and reports essentially unchanged Hit@5 on Baby (0.0282 vs 0.0283), Office (0.0281 vs 0.0299), and Yelp (0.0162 vs 0.0166). The text states that this ablation demonstrates 'recommendation relevance' gains, but the table does not support a ranking-relevance benefit; the component's value is presumably computational. Please report latency or search-cost measurements to justify the decoder design, and avoid attributing ranking gains to it.
minor comments (5)
  1. [2.1.2, Eq. (3)] Equation (3) writes the FSQ output as an element of {0,...,L-1}^{d_fsq}, but the text and implementation describe scalar codebook indices with total vocabulary size L^{d_fsq}. Please clarify how the vector output is mapped to the token index used in the R^{15360 x 768} token embedding table.
  2. [2.2 and 2.4] The text refers to 'Fig. 1 (ii)' and 'Fig. 1 (iii)' when describing the architecture, but the architecture diagram is Figure 2; Figure 1 contains only the motivational chart. Please correct the cross-references.
  3. [Appendix 6.1, Table 4] The reported pre-training statistics for Beauty (1,464 users, 6,570 items, 13,679 interactions) and Magazine Subscriptions (380 users, 865 items) are orders of magnitude smaller than typical Amazon benchmark splits and than the other listed categories. Please verify these numbers and clarify the filtering procedure.
  4. [3.3, Fig. 3] The industrial dataset results are shown only as bar charts without numeric values or error bars. Please report the exact Hit@5 and NDCG@5 values in a table, and clarify whether the 'deployment' claim refers to an online A/B test or an offline evaluation on proprietary data.
  5. [3.7, Figs. 6-7] Figures 6 and 7 present comparisons with pre-trained baselines without numerical values or repeated-run variance. Adding a companion table with exact values and standard deviations would make the claimed advantages verifiable.

Circularity Check

1 steps flagged · score 6.0 of 10

One circular step: the scaling-law 'prediction' of the full model is a fit through the 100% data point; the zero-shot core is externally evaluated and non-circular.

  1. fitted input called prediction [Section 3.6, 'LLM-Aligned Power-Law Scaling Properties' (Figure 5)]
    "We trained five variants of RecGPT using different proportions of the full training dataset (5%, 10%, 25%, 50%, and 100%), while maintaining identical model architecture and hyperparameters. ... By fitting a power-law curve to the evaluation losses of our model variants (Figure 5), we can accurately predict the performance of the full model (indicated by ⋆)."

    The 100%-data RecGPT is one of the five variants whose evaluation losses are used to fit the power-law curve and are plotted in Figure 5. 'Predicting the performance of the full model' therefore means evaluating the fitted curve at the 100% token count, which is already an input to the fit. The marked star is a regression/interpolation through an observed point, not an out-of-sample extrapolation. The claim that the scaling law 'confirms' the full model's behavior is thus true by construction of the fitting procedure, since the target point was included among the fitting data.

full rationale

The central zero-shot claim is not circular: RecGPT is pre-trained on eleven Amazon categories, and the headline evaluations are performed on held-out Amazon categories (Baby, Games, Office) and on different platforms (Yelp, Washington, Steam) that are not part of pre-training; the industrial dataset is also external. The architecture components (FSQ, hybrid attention, Trie-constrained beam search) are compared against external baselines, and the ablation study removes components rather than importing the result from a self-citation. Self-citations appear (e.g., [2], [9], [48], [50]) but none of them is load-bearing for the zero-shot claim. The one genuine circular step is in Section 3.6: the power-law fit uses the 5%, 10%, 25%, 50%, and 100% variants, and the same 100% point is then presented as 'predicted' by the fitted curve. That prediction reduces to interpolation through an input point. This does not taint the external zero-shot evaluation, but it inflates the scaling-law section's evidential status. The inconsistent Table 1/Table 2 cells noted in the skeptic brief are a correctness/significance concern, not a circularity concern, and are not counted here.

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

The zero-shot claim depends on item text carrying enough information to replace learned IDs, and on interaction patterns learned from Amazon categories transferring to unrelated platforms. These are empirical assumptions the paper does not independently validate: no analysis shows text embeddings align with user preference similarity, and no ablation varies the text encoder or checks what happens when titles are missing. The only invented abstraction is the shared FSQ codebook, which is an internal model component with no external falsifiable handle.

free parameters (5)
  • K = 4 tokens per item = 4
    Section 6.3: each item's embedding is partitioned into four segments; chosen by hand and not ablated.
  • FSQ levels L = [8,8,8,6,5] = [8,8,8,6,5]
    Section 6.3: codebook size 15,360; the choice of quantization levels is a hyperparameter not justified by analysis.
  • Sequence length and hidden size = T=1024, d_ar=768
    Section 6.3: model capacity settings; standard values but still hand-set.
  • Decoder depth = 3 GPT-2 layers
    Section 6.3: decoder layers; no depth ablation is reported.
  • Beam search width
    Section 2.4.1 says 'top-n' beam search but n is not reported anywhere in the paper.
assumptions (4)
  • domain assumption Textual item descriptions contain enough semantic signal for zero-shot recommendation.
    Section 2.1.1 states items are represented exclusively by text; no experiment measures how much preference signal is lost when IDs are removed.
  • domain assumption MPNet embeddings are domain-agnostic and recommendation-relevant.
    Section 2.1.1 imports MPNet as a text encoder; the paper assumes semantic similarity in language space corresponds to behavioral similarity in recommendation space.
  • domain assumption Pre-training on 11 Amazon categories transfers to Yelp, Washington, Steam, and a news platform.
    Appendix 6.1 lists pre-training categories; Section 3.2 assumes universal interaction patterns across platforms without measuring platform-specific drift.
  • domain assumption Each item maps to a deterministic unique FSQ token sequence, so trie prefix constraints yield valid catalog items.
    Section 2.4.2 builds a trie over item token sequences; if two items share all tokens, beam search cannot distinguish them, and the paper does not report token collision rates.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RecGPT: A Foundation Model for Sequential Recommendation." pith.science (2026). https://pith.science/paper/RMF6RQRP

@misc{pith2026250606270,
  author       = {Pith},
  title        = {Pith review of: RecGPT: A Foundation Model for Sequential Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RMF6RQRP}},
  note         = {Machine review of arXiv:2506.06270}
}
read the original abstract

This work addresses a fundamental barrier in recommender systems: the inability to generalize across domains without extensive retraining. Traditional ID-based approaches fail entirely in cold-start and cross-domain scenarios where new users or items lack sufficient interaction history. Inspired by foundation models' cross-domain success, we develop a foundation model for sequential recommendation that achieves genuine zero-shot generalization capabilities. Our approach fundamentally departs from existing ID-based methods by deriving item representations exclusively from textual features. This enables immediate embedding of any new item without model retraining. We introduce unified item tokenization with Finite Scalar Quantization that transforms heterogeneous textual descriptions into standardized discrete tokens. This eliminates domain barriers that plague existing systems. Additionally, the framework features hybrid bidirectional-causal attention that captures both intra-item token coherence and inter-item sequential dependencies. An efficient catalog-aware beam search decoder enables real-time token-to-item mapping. Unlike conventional approaches confined to their training domains, RecGPT naturally bridges diverse recommendation contexts through its domain-invariant tokenization mechanism. Comprehensive evaluations across six datasets and industrial scenarios demonstrate consistent performance advantages.

Figures

Figures reproduced from arXiv: 2506.06270 by the authors.

Figure 1
Figure 1. Our RecGPT model demonstrates strong cross-domain zero-shot generalization capabilities, con￾sistently outperforming existing recommender systems in few-shot scenarios (even when those systems incor￾porate 10%-50% of downstream unseen data) without requiring any domain-specific training data. Inspired by the remarkable success of foun￾dation models in visual and language do￾mains [7, 8], which achieve exceptional cr… view at source ↗
Figure 2
Figure 2. Architecture of our proposed foundation model-RecGPT for recommender systems, featur [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Performance on Industrial Dataset. Zero-shot comparison of RecGPT against baselines on a [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Performance w.r.t. the volume of training data. In this section, we investigate scaling laws for recommendation foundation models. Understanding how RecGPT’s performance scales with training data reveals opportunities for resource-efficient 10 [PITH_FULL_IMAGE:figures…
Figure 5
Figure 5. Figure 5: The scaling law of RecGPT. recommender systems while providing insights into cross-domain knowledge transfer. These scal￾ing properties offer practitioners a framework for optimizing performance-computation trade-offs, potentially transforming how industry deploys reco…
Figure 6
Figure 6. Figure 6: Performance comparison of pre-trained sequential recommenders across different datasets, [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Comparison isolating architectural benefits from data scale. RecGPT-10% outperforms [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 34 canonical work pages

  1. [1]

    Time to shop for valentine’s day: Shopping occasions and sequential recommendation in e-commerce

    Jianling Wang, Raphael Louca, Diane Hu, Caitlin Cellier, James Caverlee, and Liangjie Hong. Time to shop for valentine’s day: Shopping occasions and sequential recommendation in e-commerce. In WSDM, pages 645–653, 2020

    work page 2020

  2. [2]

    Diffmm: Multi-modal diffusion model for recommendation

    Yangqin Jiang, Lianghao Xia, Wei Wei, Da Luo, Kangyi Lin, and Chao Huang. Diffmm: Multi-modal diffusion model for recommendation. In MM, 2024

    work page 2024

  3. [3]

    A matrix factorization technique with trust propagation for recommendation in social networks

    Mohsen Jamali and Martin Ester. A matrix factorization technique with trust propagation for recommendation in social networks. In Recsys, pages 135–142, 2010

    work page 2010

  4. [4]

    Deep learning for sequential recommendation: Algorithms, influential factors, and evaluations

    Hui Fang, Danning Zhang, Yiheng Shu, and Guibing Guo. Deep learning for sequential recommendation: Algorithms, influential factors, and evaluations. ACM Transactions on Information Systems (TOIS), 39(1):1–42, 2020

    work page 2020

  5. [5]

    Cross-domain recom- mendation via user interest alignment

    Chuang Zhao, Hongke Zhao, Ming He, Jian Zhang, and Jianping Fan. Cross-domain recom- mendation via user interest alignment. In WWW, pages 887–896, 2023

    work page 2023

  6. [6]

    Cold-start recommendation towards the era of large language models (llms): A comprehensive survey and roadmap

    Weizhi Zhang, Yuanchen Bei, Liangwei Yang, Henry Peng Zou, Peilin Zhou, Aiwei Liu, Yinghui Li, Hao Chen, Wang, et al. Cold-start recommendation towards the era of large language models (llms): A comprehensive survey and roadmap. arXiv preprint arXiv:2501.01945, 2025

    arXiv 2025

  7. [7]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. In NeurIPS, 2023

    2023

  8. [8]

    Deepseek-v3 technical report, 2024

    DeepSeek-AI. Deepseek-v3 technical report, 2024. URL https://arxiv.org/abs/2412. 19437

    2024

Show all 50 references
  1. [9]

    Reclm: Recommendation instruction tuning

    Yangqin Jiang, Yuhao Yang, Lianghao Xia, Da Luo, Kangyi Lin, and Chao Huang. Reclm: Recommendation instruction tuning. arXiv preprint arXiv:2412.19302, 2024

    2024 arXiv

  2. [10]

    Mpnet: Masked and permuted pre-training for language understanding

    Kaitao Song, Xu Tan, Tao Qin, Jianfeng Lu, and Tie-Yan Liu. Mpnet: Masked and permuted pre-training for language understanding. NeurIPS, 33:16857–16867, 2020

    2020

  3. [11]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018

    2018 arXiv

  4. [12]

    Xlnet: Generalized autoregressive pretraining for language understanding

    Zhilin Yang, Zihang Dai, Yiming Yang, Jaime Carbonell, Russ R Salakhutdinov, and Quoc V Le. Xlnet: Generalized autoregressive pretraining for language understanding. NeurIPS, 32, 2019

    2019

  5. [13]

    Taming transformers for high-resolution image synthesis

    Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. In CVPR, pages 12873–12883, 2021

    2021

  6. [14]

    Finite scalar quantization: Vq-vae made simple

    Fabian Mentzer, David Minnen, Eirikur Agustsson, and Michael Tschannen. Finite scalar quantization: Vq-vae made simple. arXiv preprint arXiv:2309.15505, 2023

    2023 arXiv

  7. [15]

    Language models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. NeurIPS, 33:1877–1901, 2020

    1901

  8. [16]

    Visual autoregressive modeling: Scalable image generation via next-scale prediction

    Keyu Tian, Yi Jiang, Zehuan Yuan, Bingyue Peng, and Liwei Wang. Visual autoregressive modeling: Scalable image generation via next-scale prediction. NeurIPS, 2024

    2024

  9. [17]

    Unified-io: A unified model for vision, language, and multi-modal tasks

    Jiasen Lu, Christopher Clark, Rowan Zellers, Roozbeh Mottaghi, and Aniruddha Kembhavi. Unified-io: A unified model for vision, language, and multi-modal tasks. In ICLR, 2022

    2022

  10. [18]

    Contrastive learning for representation degeneration problem in sequential recommendation

    Ruihong Qiu, Zi Huang, Hongzhi Yin, and Zijian Wang. Contrastive learning for representation degeneration problem in sequential recommendation. In WSDM, pages 813–823, 2022

    2022

  11. [19]

    Emergent abilities of large language models

    Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, et al. Emergent abilities of large language models. Transactions on Machine Learning Research (TMLR), 2022. 14

    2022

  12. [20]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timo- thée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  13. [21]

    S3-rec: Self-supervised learning for sequential recommendation with mutual information maximization

    Kun Zhou, Hui Wang, Wayne Xin Zhao, Yutao Zhu, Sirui Wang, Fuzheng Zhang, Zhongyuan Wang, and Ji-Rong Wen. S3-rec: Self-supervised learning for sequential recommendation with mutual information maximization. In CIKM, pages 1893–1902, 2020

    1902

  14. [22]

    Towards universal sequence representation learning for recommender systems

    Yupeng Hou, Shanlei Mu, Wayne Xin Zhao, Yaliang Li, Bolin Ding, and Ji-Rong Wen. Towards universal sequence representation learning for recommender systems. In KDD, pages 585–593, 2022

    2022

  15. [23]

    Learning vector-quantized item representation for transferable sequential recommenders

    Yupeng Hou, Zhankui He, Julian McAuley, and Wayne Xin Zhao. Learning vector-quantized item representation for transferable sequential recommenders. In WWW, pages 1162–1171, 2023

    2023

  16. [24]

    Recommender systems with generative retrieval

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Tran, Jonah Samost, et al. Recommender systems with generative retrieval. NeurIPS, 36:10299–10315, 2023

    2023

  17. [25]

    Text is all you need: Learning language representations for sequential recommendation

    Jiacheng Li, Ming Wang, Jin Li, Jinmiao Fu, Xin Shen, Jingbo Shang, and Julian McAuley. Text is all you need: Learning language representations for sequential recommendation. In KDD, pages 1258–1267, 2023

    2023

  18. [26]

    Idgenrec: Llm-recsys alignment with textual id learning

    Juntao Tan, Shuyuan Xu, Wenyue Hua, Yingqiang Ge, Zelong Li, and Yongfeng Zhang. Idgenrec: Llm-recsys alignment with textual id learning. In SIGIR, pages 355–364, 2024

    2024

  19. [27]

    Self-attentive sequential recommendation

    Wang-Cheng Kang and Julian McAuley. Self-attentive sequential recommendation. In ICDM, pages 197–206. IEEE, 2018

    2018

  20. [28]

    Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer

    Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang. Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer. In CIKM, pages 1441–1450, 2019

    2019

  21. [29]

    Where to go next for recommender systems? id-vs

    Zheng Yuan, Fajie Yuan, Yu Song, Youhua Li, Junchen Fu, Fei Yang, Yunzhu Pan, and Yongxin Ni. Where to go next for recommender systems? id-vs. modality-based recommender models revisited. In SIGIR, pages 2639–2649, 2023

    2023

  22. [30]

    Recommendation as language processing (rlp): A unified pretrain, personalized prompt & predict paradigm (p5)

    Shijie Geng, Shuchang Liu, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. Recommendation as language processing (rlp): A unified pretrain, personalized prompt & predict paradigm (p5). In Recsys, pages 299–315, 2022

    2022

  23. [31]

    Recommendation as instruction following: A large language model empowered recommendation approach

    Junjie Zhang, Ruobing Xie, Yupeng Hou, Wayne Xin Zhao, Leyu Lin, and Ji-Rong Wen. Recommendation as instruction following: A large language model empowered recommendation approach. Transactions on Information Systems (TOIS), 2023

    2023

  24. [32]

    Neural discrete representation learning

    Aaron Van Den Oord, Oriol Vinyals, et al. Neural discrete representation learning. NeurIPS, 30, 2017

    2017

  25. [33]

    Language models are unsupervised multitask learners

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019

    2019

  26. [34]

    Generative sequential recommendation with gptrec

    Aleksandr V Petrov and Craig Macdonald. Generative sequential recommendation with gptrec. Gen-IR@SIGIR2023 Workshop, 2023

    2023

  27. [35]

    Enhanced generative recommendation via content and collaboration integration

    Yidan Wang, Zhaochun Ren, Weiwei Sun, Jiyuan Yang, Zhixiang Liang, Xin Chen, Ruobing Xie, Su Yan, Xu Zhang, Pengjie Ren, et al. Enhanced generative recommendation via content and collaboration integration. arXiv preprint arXiv:2403.18480, 2024

    2024 arXiv

  28. [36]

    Cost: Contrastive quantization based semantic tokenization for generative recommendation

    Jieming Zhu, Mengqun Jin, Qijiong Liu, Zexuan Qiu, Zhenhua Dong, and Xiu Li. Cost: Contrastive quantization based semantic tokenization for generative recommendation. InRecsys, pages 969–974, 2024

    2024

  29. [37]

    Learnable tokenizer for llm-based generative recommendation

    Wenjie Wang, Honghui Bao, Xinyu Lin, Jizhi Zhang, Yongqi Li, Fuli Feng, See-Kiong Ng, and Tat-Seng Chua. Learnable tokenizer for llm-based generative recommendation. In CIKM, 2024. 15

    2024

  30. [38]

    End-to-end learnable item tokenization for generative recommendation

    Enze Liu, Bowen Zheng, Cheng Ling, Lantao Hu, Han Li, and Wayne Xin Zhao. End-to-end learnable item tokenization for generative recommendation. SIGIR, 2025

    2025

  31. [39]

    Bridging language and items for retrieval and recommendation

    Yupeng Hou, Jiacheng Li, Zhankui He, An Yan, Xiusi Chen, and Julian McAuley. Bridging language and items for retrieval and recommendation. arXiv preprint arXiv:2403.03952, 2024

    2024 arXiv

  32. [40]

    Uctopic: Unsupervised contrastive learning for phrase representations and topic mining

    Jiacheng Li, Jingbo Shang, and Julian McAuley. Uctopic: Unsupervised contrastive learning for phrase representations and topic mining. ACL, 2022

    2022

  33. [41]

    Personalized show- cases: Generating multi-modal explanations for recommendations

    An Yan, Zhankui He, Jiacheng Li, Tianyang Zhang, and Julian McAuley. Personalized show- cases: Generating multi-modal explanations for recommendations. In SIGIR, pages 2251–2255, 2023

    2023

  34. [42]

    Improved recurrent neural networks for session- based recommendations

    Yong Kiam Tan, Xinxing Xu, and Yong Liu. Improved recurrent neural networks for session- based recommendations. In Proceedings of the 1st workshop on deep learning for recommender systems, pages 17–22, 2016

    2016

  35. [43]

    Parallel recurrent neural network architectures for feature-rich session-based recommendations

    Balázs Hidasi, Massimo Quadrana, Alexandros Karatzoglou, and Domonkos Tikk. Parallel recurrent neural network architectures for feature-rich session-based recommendations. In Recsys, pages 241–248, 2016

    2016

  36. [44]

    Personalized top-n sequential recommendation via convolutional sequence embedding

    Jiaxi Tang and Ke Wang. Personalized top-n sequential recommendation via convolutional sequence embedding. In WSDM, pages 565–573, 2018

    2018

  37. [45]

    Feature-level deeper self-attention network for sequential recommendation

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

    2019

  38. [46]

    Contrastive learning for sequential recommendation

    Xu Xie, Fei Sun, Zhaoyang Liu, Shiwen Wu, Jinyang Gao, Jiandong Zhang, Bolin Ding, and Bin Cui. Contrastive learning for sequential recommendation. In ICDE, pages 1259–1273. IEEE, 2022

    2022

  39. [47]

    Intent contrastive learning for sequential recommendation

    Yongjun Chen, Zhiwei Liu, Jia Li, Julian McAuley, and Caiming Xiong. Intent contrastive learning for sequential recommendation. In WWW, pages 2172–2182, 2022

    2022

  40. [48]

    Graph masked autoencoder for sequential recommendation

    Yaowen Ye, Lianghao Xia, and Chao Huang. Graph masked autoencoder for sequential recommendation. In SIGIR, pages 321–330, 2023

    2023

  41. [49]

    Recbole: Towards a unified, comprehensive and efficient framework for recommendation algorithms

    Wayne Xin Zhao, Shanlei Mu, Yupeng Hou, Zihan Lin, Yushuo Chen, Xingyu Pan, Kaiyuan Li, Yujie Lu, Hui Wang, Changxin Tian, et al. Recbole: Towards a unified, comprehensive and efficient framework for recommendation algorithms. In CIKM, pages 4653–4664, 2021

    2021

  42. [50]

    Sslrec: A self-supervised learning framework for recommendation

    Xubin Ren, Lianghao Xia, Yuhao Yang, Wei Wei, Tianle Wang, Xuheng Cai, and Chao Huang. Sslrec: A self-supervised learning framework for recommendation. In WSDM, pages 567–575, 2024. 16 6 Appendix 6.1 Datasets The comprehensive evaluation framework encompasses diverse datasets ...

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.