Pith. sign in

REVIEW 5 major objections 7 minor 38 references

One decoder-only sequence can run generative recall and multi-objective ranking together, bridging them with the SID trajectory.

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 ↗

T0 review · grok-4.5

2026-07-31 14:52 UTC pith:JQSQZ3FA

load-bearing objection Clean industrial unification of generative recall and multi-objective ranking in one decoder sequence; real online gains, but the history-discard “free lunch” is oversold by their own ablation. the 5 major comments →

arxiv 2607.24439 v1 pith:JQSQZ3FA submitted 2026-07-27 cs.IR

Unifying Generative Recall and Multi-Objective Ranking in a Single Decoder-Only Sequence

Ruochen Yang , Shuang Wen , Pengbo Xu , Yusheng Huang , Jiangxia Cao , Shuang Yang , Zhaojie Liu , Jiawei Sheng
show 1 more author
Tingwen Liu
This is my paper
classification cs.IR
keywords generative recommendationmulti-objective rankingdecoder-only Transformersemantic IDunified recall and rankingLoRAprefix-causal attentionindustrial recommender systems
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.

Industrial recommenders usually split recall and ranking into separate models, which wastes repeated user-context work, drops information at the hand-off, and optimizes mismatched objectives. This paper argues that both stages now share a Transformer backbone, so they can live in one heterogeneous sequence: user context, then the item’s semantic-ID trajectory, then item features. In that sequence the generated SID path becomes a live representation bridge that ranking can attend to, while Dual-Query Prefix-Causal Attention gives each task its own visibility and ranking-side LoRA keeps ranking adaptable without wrecking the generative backbone. Offline results on large industrial logs and long online A/B tests show gains on both retrieval and multi-objective scoring, plus lower end-to-end serving latency from cache reuse. A sympathetic reader cares because the cascade’s classic friction—objective gap, lost trajectory, duplicate encoding—is attacked at the sequence and serving level rather than with another hand-off module.

Core claim

UniR² shows that generative recall and multi-objective ranking can be formulated as one decoder-only Transformer over a single sequence of user context, SID trajectory, and item features. The SID trajectory acts as a representation bridge; Dual-Query Prefix-Causal Attention supplies task-specific visibility while base attention weights stay shared; stop-gradient plus ranking-only LoRA isolate optimization so ranking can adapt without collapsing autoregressive generation. The same forward pass and cached KV states then support both corpus-level generation and multi-objective scoring in one service.

What carries the argument

Dual-Query Prefix-Causal Attention (DQ-PCA) on the unified sequence [user context ∥ SID trajectory ∥ item features]: a generative query sees the full user prefix causally for next-SID prediction, while a ranking query sees user profile, the full SID trajectory, and item features (discarding long history) and adapts via ranking-only LoRA on Q/K/V, so representation couples forward and gradients stay isolated.

Load-bearing premise

Ranking can drop the long user-history prefix and still score well because the SID trajectory, a short profile token, and item-side features already carry enough matching signal.

What would settle it

An ablation or online slice where sparse or cold objectives (e.g., gift) degrade once long history is discarded for the ranking query, or where joint training without LoRA/stop-gradient shows recall collapsing under multi-objective BCE while ranking rises—showing the bridge or isolation failed.

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

If this is right

  • Recall and ranking can share one serving instance and reuse user-prefix plus SID-trajectory KV cache, cutting redundant context compute and end-to-end latency.
  • Ranking can condition on the full generation path rather than a discrete candidate list, narrowing the cascade semantic gap.
  • Scaling depth and width of the shared backbone can lift both SID generation quality and multi-objective AUC without a forced seesaw, under the isolation design.
  • Two-stage training (stabilize generation, then attach ranking) plus target-adaptive sample masks become a practical recipe for joint industrial training.
  • Online multi-behavior metrics (play, follow, like, gifting) can move together when one model owns both candidate generation and scoring.

Where Pith is reading between the lines

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

  • If SID trajectories are a reliable bridge, other cascade boundaries (e.g., pre-rank to rank, or multi-channel recall fusion) may also collapse into one sequence with dual-query masks rather than separate towers.
  • The history-discard choice implies generative SIDs are doing heavy interest compression; measuring how much cold-start or long-tail signal survives that compression is a natural next measurement.
  • Ranking-only LoRA on shared attention is a template for any generative-plus-discriminative pair where full multi-task gradients would otherwise destroy autoregressive structure.
  • Parallelizing strategy filters with ranking while reusing GPU cache suggests unified models win as much from systems scheduling as from pure accuracy.

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 / 7 minor

Summary. The paper proposes UniR², a single decoder-only Transformer that performs generative recall (autoregressive SID prediction) and multi-objective ranking within one heterogeneous sequence S = [P ∥ T_gen ∥ T_rank]. A Dual-Query Prefix-Causal Attention (DQ-PCA) scheme gives the generative query prefix-causal visibility over the user segment while the ranking query attends only to the user profile token, the SID trajectory, and item-feature tokens (the long user history is discarded for ranking). Optimization isolation is achieved by stop-gradient on the shared trunk plus ranking-only LoRA on detached Q/K/V projections, trained in a two-stage schedule. Offline experiments on large-scale Kuaishou live-streaming logs show recall gains over KuaiFormer/GNN/OneRec/OneLive/NEZHA/PROMISE (Table 1) and ranking gains over the production HoME model (Table 2); a two-week 5% online A/B replacing both the OneLive recall model and the production pre-ranker shows positive gains on play volume, follow/like rate, and gifting metrics. A ~54% end-to-end latency reduction is reported from KV-cache reuse and parallelizing ranking behind the strategy service.

Significance. If the results hold, this is a substantive industrial contribution: it is, to my knowledge, among the first demonstrations of generative recall and multi-objective ranking sharing one token sequence with layer-wise interaction (rather than post-hoc scoring or KV transfer), validated at real platform scale. Specific strengths worth crediting: (i) component ablations that isolate each design decision (Table 3), including an honest seesaw-collapse demonstration when DQ-PCA and stop-gradient are removed; (ii) attention visualizations corroborating the representation-bridge story (Fig. 4); (iii) depth/width scaling curves (Fig. 5); and (iv) a multi-metric, two-week online A/B on two apps. The two-stage training rationale and the target-adaptive masking for joint discriminative/generative sampling are clearly described and practically useful. The paper is empirical rather than derivation-based, and its metrics are standard external benchmarks, so the central claims are in principle falsifiable and are tested against reasonable baselines.

major comments (5)
  1. [§4.2, Eq. (9)–(10), Table 3, §6.3.1] The efficiency story depends on discarding the long user history for the ranking query, justified in §6.3.1 as 'causing little degradation in ranking performance.' This is contradicted by the paper's own Table 3: the 'w/o. history discard' variant (which retains history) improves over UniR² by +2.38% relative AUC@CTR and +2.87% relative AUC@GTR. By industrial ranking standards (where ~0.1% absolute AUC is routinely deployable), a 2–3% relative gap is a substantial quality concession traded for the FLOPs reduction (9.37G → 15.39G), not 'little degradation.' The manuscript should (a) reword the claim honestly as a quality–latency trade-off, (b) report absolute AUC values for this row rather than only relative deltas, and (c) provide a per-user-segment breakdown (e.g., by history length / activity level) and per-objective breakdown. The condensation assumption (SID trajectory + GSU tokens c
  2. [§4.4 Eq. (17) vs. §5.2 Eq. (19)] There is a technical inconsistency between ranking-side LoRA and the KV-cache reuse that underpins the latency claim. Eq. (17) applies low-rank residuals to X ∈ {Q, K, V} for the ranking view. But §5.2 states the ranking query 'directly attend[s] to the cached user context and the corresponding SID trajectory without recomputing either segment,' where the cache C_{u,v} (Eq. 19) stores K/V computed during recall with the base projections W_k, W_v. If the ranking view uses LoRA-adapted K/V projections (sg(W_X) + ΔW_X), the cached K/V from the recall pass cannot be reused as-is — either the K/V of P and T_gen must be recomputed with the LoRA path (reducing the claimed cache savings), or the LoRA is in effect only on the query side (in which case Eq. (17)'s X ∈ {K, V} is inaccurate). The manuscript must clarify exactly which projections are adapted, what is recomputed at ranking time, and ho
  3. [Table 2 vs. Table 3] The reported AUCs are inconsistent across tables. Table 2 reports UniR² AUC@CTR = 0.8513 and AUC@GTR = 0.9556 against Base HoME at 0.8450/0.9541. Table 3 reports UniR² AUC@CTR = 0.8450 and AUC@GTR = 0.9541 — numerically identical to Table 2's *baseline*. Either Table 3 uses a different (smaller) UniR² configuration that is not disclosed, or a row was copied incorrectly. Since Table 3 is the load-bearing evidence for all component claims, the experimental configuration underlying it (depth, width, training schedule, dataset window) and its relationship to Table 2's model must be stated explicitly.
  4. [Fig. 6, §6.3.4, §5.2] The headline -54.29% inference-time figure aggregates two separable sources: (i) unification (shared user-context encoding, KV reuse) and (ii) dispatching ranking in parallel behind the strategy-filtering service. Source (ii) is a pipeline-scheduling optimization that is largely orthogonal to model unification — a cascaded recall+ranking system could in principle apply the same overlap. As presented, the figure does not decompose the savings, so the reader cannot tell how much of the 54% is attributable to the architectural contribution. Please report the latency breakdown with and without the strategy-overlap parallelism, ideally for both the baseline pipeline and UniR².
  5. [§6.1.3, §6.2, §6.4] Two baseline-related gaps weaken the empirical positioning. First, the related-work section (§2.3) discusses unified recall–ranking systems (OnePiece, OneRanker, UniPinRec), yet none appears as an experimental baseline; the only ranking baseline is the production HoME model. At minimum, the paper should discuss why these are not comparable (e.g., different scenario, unavailable) or include one. Second, there is a mismatch between offline and online claims: Table 2 compares against the production *ranking* model, but §6.4 states UniR² replaced the recall model and the production *pre-ranking* model. The online treatment thus does not test UniR² against the full ranking stack evaluated offline, and the manuscript should state precisely which stages were substituted and what the downstream ranker consumed.
minor comments (7)
  1. [§4.4, Eq. (14)] Notation inconsistency: the text says 'we also concatenate the output at the last SID position h^rank_L,' but Eq. (14) uses h^gen_L. Since the last SID position belongs to the generative segment, presumably h^gen_L is intended; please align text and equation and define the indexing.
  2. [§6.4] The online A/B reports point estimates (+1.177% play volume, etc.) without confidence intervals or significance testing, and the treatment metric set differs between the two apps (play/follow/like on Kuaishou vs. gifting metrics on Lite). Please report statistical significance and explain the asymmetric metric choice.
  3. [Fig. 5] The y-axis 'Improvement (%)' does not state the reference configuration (e.g., relative to 2 layers / 128 hidden dim?), and the x-axis of (b) mixes the chosen operating point (640) without marking it. Please specify the baseline for the relative improvements.
  4. [§6.1.2] ACC is listed as a recall metric and ACC@all appears in Table 3, but its definition (token-level accuracy over all SID layers? over all positions?) is never given. HR@k and MRR@k for generative recall should also specify how beam-search outputs are mapped to hit evaluation.
  5. [§3, §6.1.1] Experimental protocol details are thin: dataset time window, train/test split (chronological?), number of evaluation samples, and whether Tables 1–3 use identical splits are not reported. The codebook size 8129 is unusual (not 8192); a one-line justification would help reproducibility.
  6. [§4.2, Abstract] Minor language issues: 'the two target segments play roles' (missing 'different'); 'unified integration,.' (stray punctuation in Introduction); 'serves as the basic of our platform' → 'basis'; 'two-stage different optimization objectives' in §1 is garbled. A proofreading pass is needed throughout.
  7. [§5.1] The downsampling ratio for the generative branch ('keeps all clicked samples and randomly exposed samples') is not quantified, and the NTP level weights α_i and objective weights ω_t are never instantiated, even though they are load-bearing hyperparameters for reproducing the joint-training balance.

Circularity Check

0 steps flagged

Empirical architecture paper: losses and metrics are external benchmarks, not identities forced by the equations.

full rationale

UniR² is a systems/architecture contribution (heterogeneous decoder-only sequence, DQ-PCA, ranking-side LoRA, two-stage training, single-service KV reuse). Its load-bearing claims are empirical: HR/MRR on generative recall, AUC/UAUC on multi-objective ranking, end-to-end latency breakdown, and online A/B business KPIs on Kuaishou. The training objectives (weighted NTP on SIDs; multi-objective BCE) and evaluation metrics are standard external benchmarks; nothing in Eqs. 1–19 equates a reported metric to a fitted input by construction. Self-citations to OneRec/OneLive/PROMISE/NEZHA and the internal HoME ranker supply baselines and backbone lineage, which is normal industrial practice and does not force UniR²’s gains. Ablations (Table 3) and scaling plots further treat components as falsifiable design choices, not definitional identities. No self-definitional loop, fitted-input-as-prediction, uniqueness import, or renamed known law is present. Circularity score is therefore 0.

Axiom & Free-Parameter Ledger

7 free parameters · 6 axioms · 3 invented entities

Load-bearing content is mostly engineering assumptions and design choices rather than mathematical axioms. The central claim rests on SID quantization fidelity, sufficiency of dual-query visibility, effectiveness of stop-gradient+LoRA isolation, two-stage curriculum, and industrial sampling masks—plus many fitted hyperparameters (depth, width, LoRA rank, loss weights).

free parameters (7)
  • Transformer depth N = 3 (online)
    Chosen via scaling sweep; online service uses N=3 as performance/resource tradeoff (§6.3.3).
  • Hidden dimension d = 640 (online)
    Width scaling sweep; online uses 640 (§6.3.3).
  • SID layers L and codebook size = L=3, codebook 8129
    Res-Kmeans quantization hyperparameters fixed for the scenario (§3).
  • NTP level weights α_i = α1 ≥ … ≥ αL > 0 (unspecified numeric)
    Hand-set decreasing weights favoring coarser SID codes in L_gen (Eq. 13).
  • Multi-objective BCE weights ω_t
    Task weights for CTR/LVTR/GTR etc. in L_rank (Eq. 15).
  • LoRA rank r and ranking LoRA matrices
    Low-rank adapters on ranking Q/K/V paths; rank and init are design knobs (§4.4).
  • Generative downsampling / target-adaptive ranker masks
    Sampling policy balancing exposure-distribution fitting vs discriminative negatives (§5.1); directly affects both losses.
axioms (6)
  • domain assumption Hierarchical SID tokens from discrete quantization are a sufficient generative interface to a multi-million item corpus for industrial recall.
    Inherited from TIGER/OneRec-style generative recommendation (§2.1, §3); UniR² does not re-derive tokenizer quality.
  • domain assumption Causal next-token prediction on SIDs plus beam search is an appropriate recall objective, while multi-objective BCE on XTR labels is appropriate for ranking.
    Standard industrial losses restated in §3 Eqs. 1–2 and §4.3–4.4.
  • ad hoc to paper Ranking does not need full bidirectional attention over long user history if it can attend user profile + SID trajectory + item features (history discard).
    Core efficiency assumption of DQ-PCA ranking query (§4.2 Eqs. 9–10); supported mainly by internal ablation.
  • ad hoc to paper Stop-gradient on shared trunk plus ranking-only LoRA isolates optimization enough to avoid destructive seesaw between NTP and multi-task BCE.
    Stated motivation in §4.4; ablation without LoRA/sg shows large swings, but long-horizon stability is assumed.
  • ad hoc to paper Two-stage training (generative warm-start then joint ranker) is required because early unstable SIDs would mislead the ranker.
    Curriculum prior in §5.1; not exhaustively compared to fully joint from-scratch alternatives in the text.
  • standard math Standard multi-head attention and Transformer block algebra.
    Used throughout §4.2 Eq. 11.
invented entities (3)
  • UniR² unified heterogeneous decoder sequence (P ∥ T_gen ∥ T_rank) no independent evidence
    purpose: Single forward trunk coupling user context, SID trajectory, and item features for both tasks.
    Architectural object defined in §4.1; evaluated only inside this paper’s stack.
  • Dual-Query Prefix-Causal Attention (DQ-PCA) no independent evidence
    purpose: Give generative and ranking queries different visibility masks while sharing base W_qkv.
    Named mechanism in §4.2; ablations compare variants but no external independent validation.
  • Ranking-side LoRA on detached shared projections no independent evidence
    purpose: Restore ranking feature-fusion plasticity under stop-gradient isolation.
    Adaptation of LoRA [9] to this dual-task setting (§4.4); evidence is internal ablation/attention plots.

pith-pipeline@v1.2.0-grok45-kimik3 · 20946 in / 4105 out tokens · 82892 ms · 2026-07-31T14:52:58.133811+00:00 · methodology

0 comments
read the original abstract

Modern industrial recommendation systems typically separate recall and ranking into two independent stages. Although this cascade supports corpus-level retrieval and fine-grained multi-objective scoring, it causes objective inconsistency, information loss at the candidate hand-off, and redundant user-side context computation. Meanwhile, the generative recall and ranking scaling share a common Transformer-based modeling philosophy, where architectural consistency creates a natural opportunity for unified integration. However, direct sharing remains challenging since the two tasks require different information visibility and optimization methods. Therefore, we propose \textbf{UniR$^2$}, a \textbf{Uni}fied decoder-only Transformer that unifies Generative \textbf{R}ecall and Multi-Objective \textbf{R}anking within a single heterogeneous sequence comprising user context, SID trajectory, and item features. Within this sequence, the generated trajectory serves as a representation bridge between recall and ranking, where Dual-Query Prefix-Causal Attention provides task-specific visibility. The two tasks share the base attention weights but retain separate optimization boundaries, with ranking-side LoRA preserving ranking adaptability without disrupting the generative backbone. Extensive offline experiments on large-scale industrial data demonstrate the effectiveness and efficiency of UniR$^2$ for both recall and ranking. Long-term online A/B tests on Kuaishou platform further show consistent positive gains, validating the practicality of unified model in large-scale recommendation systems.

Figures

Figures reproduced from arXiv: 2607.24439 by Jiangxia Cao, Jiawei Sheng, Pengbo Xu, Ruochen Yang, Shuang Wen, Shuang Yang, Tingwen Liu, Yusheng Huang, Zhaojie Liu.

Figure 1
Figure 1. Figure 1: Architecture comparison between cascaded architec [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The overall architecture of our model. Since P represents stable user side information and remains un￾changed across different items, it serves as a shared memory read by each downstream target token, playing the role of an encoder key/value cache. Generative segment Tgen. Given the target item 𝑣 with hierar￾chical SID 𝑠𝑣 = (𝑞1, 𝑞2, 𝑞3), this segment is the teacher-forced input of the generative recall tas… view at source ↗
Figure 3
Figure 3. Figure 3: The inference pipeline of UniR2 . GPU caches the user context and SID-trajectory KV states computed during recall for direct reuse in ranking, while strategy filtering and ranking are executed in parallel. ranking processes, the generated candidates should undergo fur￾ther processing through commercial strategies such as qualification review, deduplication, and content filtering before being passed on to t… view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of attention weights of recall and [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: As can be observed, scaling from both brings substantial [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The comparison of system inference time. [PITH_FULL_IMAGE:figures/full_fig_p009_6.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

38 extracted references · 10 linked inside Pith

  1. [1]

    Zheng Chai, Qin Ren, Xijun Xiao, Huizhi Yang, Bo Han, Sijun Zhang, Di Chen, Hui Lu, Wenlin Zhao, Lele Yu, et al . 2025. Longer: Scaling up long sequence modeling in industrial recommenders. InProceedings of the Nineteenth ACM Conference on Recommender Systems. 247–256

    2025

  2. [2]

    Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. InProceedings of the 10th ACM conference on recommender systems. 191–198

    2016

  3. [3]

    Sunhao Dai, Jiakai Tang, Jiahua Wu, Kun Wang, Yuxuan Zhu, Bingjun Chen, Bangyang Hong, Yu Zhao, Cong Fu, Kangle Wu, et al. 2025. Onepiece: Bringing context engineering and reasoning to industrial cascade ranking system.arXiv preprint arXiv:2509.18091(2025)

    arXiv 2025

  4. [4]

    Jiaxin Deng, Shiyao Wang, Kuo Cai, Lejian Ren, Qigen Hu, Weifeng Ding, Qiang Luo, and Guorui Zhou. 2025. Onerec: Unifying retrieve and rank with generative recommender and iterative preference alignment.arXiv preprint arXiv:2502.18965 (2025)

    Pith/arXiv arXiv 2025

  5. [5]

    Haiwen Diao, Penghao Wu, Hanming Deng, Jiahao Wang, Shihao Bai, Silei Wu, Weichen Fan, Wenjie Ye, Wenwen Tong, Xiangyu Fan, et al. 2026. Sensenova-u1: Conference acronym ’XX, June 03–05, 2018, Woodstock, NY Ruochen Yang, et al. Unifying multimodal understanding and generation with neo-unify architecture. arXiv preprint arXiv:2605.12500(2026)

    Pith/arXiv arXiv 2026

  6. [6]

    Kairui Fu, Changfa Wu, Kun Yuan, Binbin Cao, Dunxian Huang, Yuliang Yan, Junjun Zheng, Jianning Zhang, Silu Zhou, Jian Wu, et al. 2026. RankGR: Rank- Enhanced Generative Retrieval with Listwise Direct Preference Optimization in Recommendation.arXiv preprint arXiv:2602.08575(2026)

    arXiv 2026

  7. [7]

    Chengcheng Guo, Kuo Cai, Yu Zhou, Qiang Luo, Ruiming Tang, Han Li, Kun Gai, and Guorui Zhou. 2026. PROMISE: Process Reward Models Unlock Test-Time Scaling Laws in Generative Recommendations.arXiv preprint arXiv:2601.04674 (2026)

    arXiv 2026

  8. [8]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. InProceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval. 639–648

    2020

  9. [9]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, et al. 2022. Lora: Low-rank adaptation of large language models.Iclr1, 2 (2022), 3

    2022

  10. [10]

    Xu Huang, Hao Zhang, Zhifang Fan, Yunwen Huang, Zhuoxing Wei, Zheng Chai, Jinan Ni, Yuchao Zheng, and Qiwei Chen. 2026. MixFormer: Co-Scaling Up Dense and Sequence in Industrial Recommenders.arXiv preprint arXiv:2602.14110 (2026)

    Pith/arXiv arXiv 2026

  11. [11]

    Jie Jiang, Yangru Huang, Zeyu Wang, Changping Wang, Yuling Xiong, Jun Zhang, and Huan Yu. 2026. Spend Search Where It Pays: Value-Guided Structured Sampling and Optimization for Generative Recommendation.arXiv preprint arXiv:2602.10699(2026)

    arXiv 2026

  12. [12]

    Hanyu Li, Yi-Ping Hsu, Aditya Mantha, Prabhat Agarwal, Laksh Bhasin, Jialu Wang, Hongtao Lin, Bella Huang, Yaxin Li, Xinyi Li, et al . 2026. UniPinRec: Unifying Generative Retrieval and Ranking at Pinterest Scale.arXiv preprint arXiv:2606.00422(2026)

    Pith/arXiv arXiv 2026

  13. [13]

    Chi Liu, Jiangxia Cao, Rui Huang, Kai Zheng, Qiang Luo, Kun Gai, and Guorui Zhou. 2024. Kuaiformer: Transformer-based retrieval at kuaishou.arXiv preprint arXiv:2411.10057(2024)

    Pith/arXiv arXiv 2024

  14. [14]

    Yu Liu and Jiangxia Cao. 2026. Harmonizing Generative Retrieval and Ranking in Chain-of-Recommendation.arXiv preprint arXiv:2604.25787(2026)

    Pith/arXiv arXiv 2026

  15. [15]

    Xinchen Luo, Jiangxia Cao, Tianyu Sun, Jinkai Yu, Rui Huang, Wei Yuan, Hezheng Lin, Yichen Zheng, Shiyao Wang, Qigen Hu, et al . 2025. Qarm: Quantitative alignment multi-modal recommendation at kuaishou. InProceedings of the 34th ACM International Conference on Information and Knowledge Management. 5915– 5922

    2025

  16. [16]

    Jiaqi Ma, Zhe Zhao, Xinyang Yi, Jilin Chen, Lichan Hong, and Ed H Chi. 2018. Modeling task relationships in multi-task learning with multi-gate mixture-of- experts. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 1930–1939

    2018

  17. [17]

    Qi Pi, Guorui Zhou, Yujing Zhang, Zhe Wang, Lejian Ren, Ying Fan, Xiaoqiang Zhu, and Kun Gai. 2020. Search-based user interest modeling with lifelong sequential behavior data for click-through rate prediction. InProceedings of the 29th ACM International Conference on Information & Knowledge Management. 2685–2692

    2020

  18. [18]

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

  19. [19]

    Dekai Sun, Yiming Liu, Jiafan Zhou, Xun Liu, Chenchen Yu, Yi Li, Jun Zhang, Huan Yu, and Jie Jiang. 2026. OneRanker: Unified Generation and Ranking with One Model in Industrial Advertising Recommendation.arXiv preprint arXiv:2603.02999(2026)

    arXiv 2026

  20. [20]

    Yijia Sun, Shanshan Huang, Zhiyuan Guan, Qiang Luo, Ruiming Tang, Kun Gai, and Guorui Zhou. 2026. GRank: Towards Target-Aware and Streamlined Industrial Retrieval with a Generate-Rank Framework. InProceedings of the ACM Web Conference 2026. 7798–7808

    2026

  21. [21]

    Hongyan Tang, Junning Liu, Ming Zhao, and Xudong Gong. 2020. Progressive layered extraction (ple): A novel multi-task learning (mtl) model for personalized recommendations. InProceedings of the 14th ACM conference on recommender systems. 269–278

    2020

  22. [22]

    Jiakai Tang, Runfeng Zhang, Weiqiu Wang, Yifei Liu, Chuan Wang, Xu Chen, Yeqiu Yang, Jian Wu, Yuning Jiang, and Bo Zheng. 2026. LoopCTR: Unlock- ing the Loop Scaling Power for Click-Through Rate Prediction.arXiv preprint arXiv:2604.19550(2026)

    Pith/arXiv arXiv 2026

  23. [23]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need.Advances in neural information processing systems30 (2017)

    2017

  24. [24]

    Chunqi Wang, Bingchao Wu, Taotian Pang, Jiahao Wang, Jie Yang, Jia Liu, Hao Zhang, Hai Zhu, Lei Shen, Shizhun Wang, et al. 2026. SORT: A Systematically Op- timized Ranking Transformer for Industrial-scale Recommenders.arXiv preprint arXiv:2603.03988(2026)

    arXiv 2026

  25. [25]

    Ruoxi Wang, Bin Fu, Gang Fu, and Mingliang Wang. 2017. Deep & cross network for ad click predictions. InProceedings of the ADKDD’17. 1–7

    2017

  26. [26]

    Shen Wang, Yusheng Huang, Ruochen Yang, Shuang Wen, Pengbo Xu, Jiangxia Cao, Yueyang Liu, Kuo Cai, Chengcheng Guo, Shiyao Wang, et al. 2026. OneLive: Dynamically Unified Generative Framework for Live-Streaming Recommenda- tion.arXiv preprint arXiv:2602.08612(2026)

    arXiv 2026

  27. [27]

    Xu Wang, Jiangxia Cao, Zhiyi Fu, Kun Gai, and Guorui Zhou. 2025. Home: Hierarchy of multi-gate experts for multi-task learning at kuaishou. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    2025

  28. [28]

    Yejing Wang, Shengyu Zhou, Jinyu Lu, Ziwei Liu, Langming Liu, Maolin Wang, Wenlin Zhang, Feng Li, Wenbo Su, Pengjie Wang, et al. 2026. Nezha: A zero- sacrifice and hyperspeed decoding architecture for generative recommendations. InProceedings of the ACM Web Conference 2026. 8073–8082

    2026

  29. [29]

    Zhipeng Wei, Kuo Cai, Junda She, Jie Chen, Minghao Chen, Yang Zeng, Qiang Luo, Wencong Zeng, Ruiming Tang, Kun Gai, et al. 2026. Oneloc: Geo-aware generative recommender systems for local life service. InProceedings of the Nineteenth ACM International Conference on Web Search and Data Mining. 735–744

    2026

  30. [30]

    Ruochen Yang, Yueyang Liu, Zijie Zhuang, Changxin Lao, Yuhui Zhang, Jiangxia Cao, Jia Xu, Xiang Chen, Haoke Xiao, Xiangyu Wu, et al . 2026. SARM: LLM- Augmented Semantic Anchor for End-to-End Live-Streaming Ranking.arXiv preprint arXiv:2602.09401(2026)

    arXiv 2026

  31. [31]

    Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L Hamilton, and Jure Leskovec. 2018. Graph convolutional neural networks for web-scale recommender systems. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 974–983

    2018

  32. [32]

    Buyun Zhang, Liang Luo, Yuxin Chen, Jade Nie, Xi Liu, Daifeng Guo, Yanli Zhao, Shen Li, Yuchen Hao, Yantao Yao, et al. 2024. Wukong: Towards a scaling law for large-scale recommendation.arXiv preprint arXiv:2403.02545(2024)

    Pith/arXiv arXiv 2024

  33. [33]

    Kun Zhang, Jingming Zhang, Wei Cheng, Yansong Cheng, Jiaqi Zhang, Hao Lu, Xu Zhang, Haixiang Gan, Jiangxia Cao, Tenglong Wang, et al. 2026. OneMall: One Model, More Scenarios–End-to-End Generative Recommender Family at Kuaishou E-Commerce.arXiv preprint arXiv:2601.21770(2026)

    arXiv 2026

  34. [34]

    Zhaoqi Zhang, Haolei Pei, Jun Guo, Tianyu Wang, Yufei Feng, Hui Sun, Shaowei Liu, and Aixin Sun. 2026. Onetrans: Unified feature interaction and sequence modeling with one transformer in industrial recommender. InProceedings of the ACM Web Conference 2026. 8162–8170

    2026

  35. [35]

    Guorui Zhou, Hengrui Hu, Hongtao Cheng, Huanjie Wang, Jiaxin Deng, Jinghao Zhang, Kuo Cai, Lejian Ren, Lu Ren, Liao Yu, et al. 2025. Onerec-v2 technical report.arXiv preprint arXiv:2508.20900(2025)

    Pith/arXiv arXiv 2025

  36. [36]

    Guorui Zhou, Xiaoqiang Zhu, Chenru Song, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. 2018. Deep interest network for click-through rate prediction. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 1059–1068

    2018

  37. [37]

    Yifeng Zhou, Yuehong Hu, Zhixiang Feng, Junwei Pan, Kaihui Wu, Hanyong Li, Shangyu Zhang, Shudong Huang, Zhangbin Zhu, Chengguo Yin, et al. 2026. TokenFormer: Unify the Multi-Field and Sequential Recommendation Worlds. arXiv preprint arXiv:2604.13737(2026)

    Pith/arXiv arXiv 2026

  38. [2023]

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

    2023