Pith. sign in

REVIEW 4 major objections 7 minor 47 references

CogRec: Structure-Cognitive Fast-and-Slow Reasoning for Generative Recommendation

T0 review · 4 major / 7 minor · reviewed 2026-07-31 · grok-4.5

Pith's one-line read Recommendation reasoning works better when it navigates the same Semantic-ID map used to name the next item.

desk verdict Solid controlled systems paper on SID-native routing; small conditional lifts, and the Direct-vs-Routing decode budgets are not matched. read the letter →

arxiv 2607.24402 v1 pith:MX5DTK5R submitted 2026-07-27 cs.IR

classification cs.IR
keywords GenerativeRecommendationSemanticIDsStructure-groundedReasoningFast-and-SlowConstrainedDecodingSIDRoutingSequential
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

Generative recommenders often turn next-item prediction into generating a hierarchical Semantic ID, but they usually treat that ID as a string to memorize, not as a map the model can travel. Natural-language “think before you recommend” steps help only loosely, because the model reasons in words and then must jump into a discrete ID space. CogRec builds a navigable SID topology—vertical hierarchy plus same-layer semantic links and item neighborhoods—and trains the model to emit layer-wise Match, LateralJump, and Explore steps before the target ID. Exact matches act as fast localization; lateral jumps and exploration are slower structural moves. Under a shared ID map and constrained decoding, this structure-grounded routing beats matched direct generation on some benchmarks, especially when history and target are not prefix-aligned but still connected by learnable SID transitions; long or weakly supported routes cost more decode steps and can accumulate errors.

What carries the argument

SID Routing: offline-constructed, layer-wise operations over an augmented Semantic-ID topology (hierarchy + intra-layer centroid graphs + item neighborhoods) that verbalize Match (fast exact code share), LateralJump (graph edge), or Explore (no edge) before trie-constrained target-SID generation, trained from a shared direct-generation checkpoint against a natural-language reasoning branch.

What would settle it

On the same split, SID map, and trie decoding, if SID Routing never beats its shared-checkpoint direct generator on Medium-difficulty cases (weak prefix overlap but short graph proximity)—or if longer routing traces systematically lower Hit@10/NDCG@10 versus direct generation across Beauty, Sports, and Toys—the conditional-usefulness claim fails.

Watch

Extended reading notes

Core claim

Grounding intermediate reasoning in the same Semantic-ID topology used for final generation—via layer-wise Match, LateralJump, and Explore (SID Routing)—improves next-item Hit@10 and NDCG@10 over an otherwise matched direct-SID generator when prefix matching is insufficient but learnable SID-space transitions remain available, while adding decode cost and error risk on long or weakly supported routes.

Load-bearing premise

That routes built offline from the true next item and hand-built similarity graphs remain a useful teaching signal at inference, when the model sees only the user history and must invent the path before naming the item.

Editorial extensions

If this is right

  • Reasoning traces for generative recommendation should be defined in the identifier space, not only as free-text rationales.
  • Systems can treat fast vs slow computation as route structure (match vs jump/explore counts and costs) without separate neural modules.
  • Gains concentrate in a middle structural regime; easy prefix matches may not need routing, and hard disconnected cases may not benefit.
  • Compact structural traces can outperform more human-readable enriched text when the decoder must emit constrained SIDs.
  • Evaluation should stratify by history–target SID proximity and route length, not only overall Hit/NDCG.

Reading between the lines

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

  • Adaptive decode policies could skip or shorten routing when predicted prefix depth is high, spending extra tokens only on medium-regime cases.
  • The same topology could support retrieval-time graph walk or beam guidance, not only supervised think-text before generation.
  • If offline Explore labels are noisy, replacing them with learned or multi-path supervision may matter more than scaling rationale length.
  • Cross-domain catalogs with unstable text embeddings may shrink the medium regime where lateral edges stay reliable.
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 / 7 minor

Summary. The paper proposes CogRec, a generative-recommendation framework in which intermediate reasoning is expressed not as free natural language but as layer-wise operations (Match / LateralJump / Explore) over a Semantic-ID topology augmented with intra-layer cosine-similarity graphs and item-level HNSW neighborhoods. Supervision routes are derived offline from ground-truth anchor–target SID relations; a multi-stage pipeline trains direct generation (Stage 2) and two reasoning branches (Stage 3a natural-language, Stage 3b SID Routing) from a shared checkpoint under the same SID mapping, candidate space, and trie-constrained decoding. On Amazon Beauty/Sports/Toys, SID Routing improves Hit@10/NDCG@10 over the matched direct checkpoint on Beauty and Sports but not Toys; difficulty-stratified analysis localizes the benefit to a "Medium" structural regime, and a progressive design study shows semantic-enriched traces and assistant-only loss masks degrade reasoning performance. Code is released.

Significance. If the central comparison holds, this is a useful and honest contribution to reasoning-enhanced generative recommendation: the controlled design (shared Stage-2 checkpoint, identical SID map/trie/candidate space, reproduced OneRec-Think branches under the same protocol) is a genuine strength rarely executed this cleanly, the Toys reversal and the Medium-regime stratification are reported without overclaiming, the progressive design study is informative, and the released code plus detailed hyperparameter table (Table 1) support reproducibility. The structural framing of "fast vs slow" as measurable route cost is a reasonable operationalization. However, the headline quantitative effect is small (ΔH@10 of +0.0012 to +0.0046), measured at a single seed, under an asymmetric decoding budget, so the practical significance currently hinges on a comparison that is not yet fully nailed down. The work is best read as a well-controlled diagnostic study of when structure-grounded reasoning helps, rather than as a new state-of-the-art method.

major comments (4)
  1. [§4.1.3, Table 1, Table 4] The controlled Direct-vs-Routing comparison is not candidate-budget-matched. The routing branch samples 5 traces (temp 1.5) and decodes beam-10 SIDs per trace, yielding 'at most 50 trace-conditioned SID hypotheses per instance' (Table 1), while direct generation uses a single beam-10 pass. Pooling 50 diverse hypotheses mechanically inflates Hit@K even if traces carry no information. The OneRec-Think reproduction under the same 5-trace protocol (Table 4, no gain on Beauty/Sports) partially mitigates this, but the clean control is a budget-matched direct baseline: e.g., 5 sampled direct generations pooled the same way, or beam-50 direct. Relatedly, the manuscript never states how the 50 hypotheses are merged into the ranked top-10 list (deduplication? ranking by model score?). Given that the headline claim rests on Table 4, this must be resolved.
  2. [Table 4, Table 6, §4.1.4] All controlled results use a single random seed (42) with no variance estimate or significance testing. The Beauty gain (ΔH@10 = +0.0012, Table 6) corresponds to roughly 27 test users out of 22,363; the Sports gain (+0.0046) is larger but still small. With per-user paired outcomes available, a McNemar test or paired bootstrap on Hit@10 between Direct and Routing is cheap and would substantially strengthen (or appropriately temper) the claim. Multi-seed Stage-3b runs would be even better. As written, the sign of the Beauty delta is not distinguishable from noise.
  3. [§4.6 (RQ4), §3.3] RQ4 analyzes routing-step distributions (Fig. 4) and generated think-lengths (Fig. 5) but never measures whether the model's generated routes are structurally faithful: Do verbalized LateralJump operations correspond to edges actually present in G^(l)? Does the generated route match the offline reference route? Is Hit@10 higher conditional on route correctness? The central mechanism claim — that grounding the trace in the SID topology steers constrained decoding, rather than merely adding autoregressive tokens — is load-bearing for the paper's thesis and is directly testable with diagnostics the authors already have. Without it, the Medium-regime improvement is consistent with several alternative explanations (e.g., extra compute tokens, ensembling).
  4. [§4.1.3 (decoding protocol)] For full-test CoT evaluation only, histories are tail-truncated to ~95th-percentile length (21/20/18 items), while direct generation and training use full histories. The Table 3 CogRec-Routing numbers are therefore computed under a different input condition than CogRec-Direct. The direction of bias is unclear (truncation plausibly hurts routing, making results conservative, but this is not argued). At minimum this asymmetry should be flagged next to Table 3; ideally, direct numbers under identical truncation, or subset-matched comparisons, should be reported.
minor comments (7)
  1. [Front matter] Template artifacts remain: CCS Concepts placeholder text ('Do Not Use This Code → Generate the Correct Terms'), 'Conference acronym 'XX', 2018 copyright block, and 'Received 20 February 2007; revised 12 March 2009'. These must be fixed before any camera-ready.
  2. [§4.1.4; Abstract] Typo: 'random sa eed 42'. Also the abstract sentence 'show that SID Routing improves its corresponding direct-generation, indicate that...' is grammatically incomplete.
  3. [Table 5 vs Table 4] Table 5's Step-0 Reason numbers (0.0895/0.0538) differ from Table 4's Beauty Routing (0.0854/0.0517) because the former uses the CoT-subset protocol. The subset is never defined (size, selection criterion). Please define it and state explicitly which protocol each table uses.
  4. [§3.2, §4.1.4] The item-level HNSW index (§3.2) is said to support 'offline topology construction and structural diagnosis,' but §4.1.4 states supervised routing labels use only exact SID matching and centroid graphs. If HNSW affects no training signal or reported diagnostic, clarify its role or trim; currently its computational cost appears unmotivated.
  5. [Table 1] Stage-3b checkpoint selection uses different epochs per dataset (Beauty 2, Sports 2, Toys 3). The paper states selection was fixed before test evaluation — good — but the validation criterion used for selection should be stated explicitly.
  6. [Figure 3] Fig. 3 normalizes color independently within each metric and panel, which exaggerates tiny absolute differences (e.g., Hard-group values near 0.01). The annotations mitigate this; consider also reporting group sizes per panel (only given in Fig. 5) and absolute-scale bars for at least one metric.
  7. [§3.3–3.4] Eq. (21): the Medium condition (M_u ≥ 1 ∨ D_u ≤ 2) makes M_u < 2 redundant in the first clause; consider restating for readability. Also report how often the fallback anchor (last item) is triggered, since it determines route origin quality.

Circularity Check

1 steps flagged · score 1.0 of 10

Empirical supervised generative recommendation; no load-bearing prediction reduces to its inputs by construction.

  1. other [§3.3–3.4 (Eq. 11–14, 20–21); RQ3 / Fig. 3; Table 6]
    "The target-dependent definitions in this subsection are used only for offline supervision construction and structural diagnostics. During inference, the model receives only the user history; the ground-truth target, anchor–target relations, and reference route are not provided. ... Using the maximum prefix depth Mu and graph distance Du, we define δu = Easy / Medium / Hard ... These labels are used for analysis and data diagnosis rather than as recommendation targets."

    Not true by-construction circularity: offline routes and difficulty labels reuse the SID topology and ground-truth target, then Medium-regime gains are partly narrated via those same labels. Evaluation metrics remain external next-item Hit/NDCG after constrained decoding, so the main claim is not forced equal to the label definition. Flagged only as mild interpretive coupling of analysis strata to the structure that defines the method.

full rationale

CogRec’s central claims are empirical Hit@K/NDCG@K comparisons under a shared SID map, Stage-2 checkpoint, and trie-constrained decoding (Tables 3–4, 6; §4.3–4.8), not closed-form first-principles predictions. Offline Match/LateralJump/Explore labels are built from ground-truth anchor–target SIDs and the hand-built intra-layer graph (§3.3, Eq. 11–14) solely as supervised training targets for Stage 3b; at inference the model sees only history, and success is scored by next-item ranking after constrained SID decoding—not by recovering the offline route. That is ordinary teacher-forced path supervision, not self-definitional circularity or a fitted parameter renamed as a prediction. Difficulty groups (Easy/Medium/Hard from Mu and Du; §3.4, Eq. 20–21) are post-hoc analysis strata, not quantities the model is trained to output or that force the main metric lifts. No uniqueness theorem or load-bearing self-citation chain underwrites the result; baselines and OneRec-Think comparisons are external or controlled reproductions. Asymmetric decode budget (5 traces × beam 10 vs direct beam 10) is a fairness/methodology concern, not circularity. Residual mild concern is only interpretive: regime claims lean on topology-defined Medium labels that reuse the same structure used to build routes—but this does not make ΔHit@10 equal the supervision by construction. Score 1 reflects that minor interpretive coupling, not a forced derivation.

Assumptions & free parameters 7 free parameters · 5 assumptions · 3 invented entities

Central empirical claim rests on standard next-item leave-one-out protocol, residual-quantized SIDs from text embeddings, hand-chosen graph and routing hyperparameters, and the modeling assumption that offline target-conditioned structural traces are good training targets for history-only inference. No physical constants; free parameters are engineering knobs that shape both supervision and the ‘middle regime’ narrative.

free parameters (7)
  • Intra-layer similarity threshold τ = 0.15
    Edges kept only if centroid cosine ≥ τ; sets which transitions are LateralJump vs Explore and thus the supervision and difficulty story.
  • Intra-layer top-k neighbors = 16
    Candidate graph degree before thresholding; chosen by authors, not derived.
  • SID depth L and codebook size K = L=4, K=256 per layer
    Defines hierarchy granularity and route length upper bound (≤4 non-Match steps).
  • Operation structural costs c(Match,LateralJump,Explore) = 0, 1, 2
    Ordinal 0/1/2 diagnostic convention for fast/slow; not learned, but frames RQ4 cost narrative.
  • Difficulty thresholds (M_u, D_u, D_max) = M≥2 easy; D_max=4
    Easy if prefix depth ≥2; Medium/Hard use first-layer BFS cap D_max=4; partitions where routing is claimed useful.
  • Stage-3b checkpoint epoch per dataset = Beauty/Sports: 2; Toys: 3
    Reported test numbers use Beauty/Sports epoch 2, Toys epoch 3 selected before test; affects claimed deltas.
  • Reasoning decode hyperparameters = 5×beam10, T=1.5/0.6
    5 traces, temp 1.5, top-p 0.95, beam 10 for SIDs; multiplies hypotheses up to 50 and affects measured Hit/NDCG.
assumptions (5)
  • domain assumption Next-item sequential recommendation under chronological leave-one-out on Amazon 2014 Beauty/Sports/Toys is a valid test of reasoning quality.
    Standard recsys protocol (§4.1); results may not transfer to industrial catalogs or multi-objective ranking.
  • domain assumption Residual K-means SIDs from BGE item-text embeddings preserve semantics useful for both generation and lateral graphs.
    Stage 0 construction (§3.2); graph edges are centroid cosine in that embedding space.
  • ad hoc to paper Offline routes from ground-truth target and anchor selection are legitimate supervised targets even though inference lacks the target.
    §3.3 explicitly target-dependent for supervision only; load-bearing for Stage 3b.
  • domain assumption Trie-constrained decoding over valid SIDs makes variants comparable in the same output space.
    §3.6 / §4.1.3; necessary for fair Direct vs Reason comparison.
  • ad hoc to paper Fast-and-slow dual-process language is only a computational lens, not a claim of human cognition.
    Stated in Intro/§3; avoids stronger cognitive axioms but still structures the method narrative.
invented entities (3)
  • Structure-cognitive SID topology (hierarchy + intra-layer graphs + item HNSW neighborhoods)
    purpose: Turn SID codes from pure output labels into a navigable reasoning space.
    Composition of known VQ SIDs and kNN graphs; packaged as the paper’s substrate for routing.
  • SID Routing operations: Match, LateralJump, Explore
    purpose: Discrete intermediate reasoning tokens aligned with SID layers before target generation.
    Defined by equality / edge membership / otherwise in Eq. (14); supervision format is paper-specific.
  • Routing cost C_u and non-Match step count s_u
    purpose: Quantify fast vs slow structural computation without separate modules.
    Diagnostic aggregates of hand-assigned operation costs; not measured latency.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CogRec: Structure-Cognitive Fast-and-Slow Reasoning for Generative Recommendation." pith.science (2026). https://pith.science/paper/MX5DTK5R

@misc{pith2026260724402,
  author       = {Pith},
  title        = {Pith review of: CogRec: Structure-Cognitive Fast-and-Slow Reasoning for Generative Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MX5DTK5R}},
  note         = {Machine review of arXiv:2607.24402}
}
read the original abstract

Semantic-ID-based generative recommendation represents each item as a hierarchical discrete token sequence and reformulates next-item prediction as constrained sequence generation. Existing methods, however, mainly use Semantic IDs as target sequences to be memorized, leaving the hierarchy, intra-layer relations, and item neighborhoods underused as an explicit reasoning space. Explicit reasoning-enhanced generative methods often produce a natural-language rationale before the item identifier, but this rationale is only weakly coupled with the discrete SID space in which the final prediction is made. We propose CogRec, a structure-cognitive fast-and-slow reasoning framework that grounds intermediate reasoning in the same SID topology used for target generation. CogRec augments the vertical SID hierarchy with intra-layer semantic graphs and item-level neighborhoods, and introduces SID Routing to represent recommendation reasoning through layer-wise Match, LateralJump, and Explore operations. Exact matching implements fast semantic localization, whereas lateral and exploratory operations instantiate slower structural navigation. A supervised multi-stage pipeline aligns the newly introduced SID tokens, establishes direct SID generation, and trains natural-language and SID-routing reasoning branches from a shared checkpoint under the same trie-constrained output space. Experiments on three public sequential-recommendation benchmarks show that SID Routing improves its corresponding direct-generation, indicate that structure-grounded reasoning is most useful when prefix matching is insufficient but learnable SID-space transitions remain available, whereas long or weakly supported routes introduce additional decoding cost and accumulated errors. Code is available at https://github.com/caskcsg/CogRec

Figures

Figures reproduced from arXiv: 2607.24402 by the authors.

Figure 1
Figure 1. Overview of CogRec. Stage 0 constructs the structure-cognitive SID topology from hierarchical Semantic IDs, intra [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Structure-cognitive SID space and SID Routing. The [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Comparison scope [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (3 more)
Figure 3
Figure 3. Figure 3: Difficulty-stratified performance of the reproduced OneRec-Think Direct/CoT branches and the CogRec Direct/Routing [PITH_FULL_IMAGE:figures/full_fig_p010_3.png]
Figure 4
Figure 4. Figure 4: Routing-step distribution across datasets. A step denotes one non- [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Think-length distribution under natural-language reasoning and SID Routing. SID Routing produces longer traces [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 1 canonical work pages

  1. [1]

    Shiteng Cao, Kaian Jiang, Yunlong Gong, and Zhiheng Li. 2026. TwiSTAR: Think Fast, Think Slow, Then Act, Generative Recommendation with Adaptive Reasoning. arXiv:2605.11553 [cs.IR] https://arxiv.org/abs/2605.11553

  2. [2]

    Wenzhuo Cheng, Menghang Gong, Qixin Guo, Hang Zheng, Zhaobin Yang, Jianguo Lou, and Zhengwei Zheng. 2026. CapsID: Soft-Routed Variable-Length Semantic IDs for Generative Recommendation. arXiv:2605.05096 [cs.IR] https: //arxiv.org/abs/2605.05096

  3. [3]

    Stephen Chung, Wenyu Du, and Jie Fu. 2025. Thinker: Learning to Think Fast and Slow. InAdvances in Neural Information Process- ing Systems, Vol. 38. Curran Associates, Inc., Red Hook, NY, USA, 102812–102841. https://proceedings.neurips.cc/paper_files/paper/2025/hash/ 94dc604e115237a7f4a758b3146cd976-Abstract-Conference.html

  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:2502.18965 [cs.IR] https://arxiv.org/abs/2502.18965

  5. [5]

    Jose, and Zhaochun Ren

    Junchen Fu, Xuri Ge, Alexandros Karatzoglou, Ioannis Arapakis, Suzan Verberne, Joemon M. Jose, and Zhaochun Ren. 2026. Differentiable Semantic ID for Genera- tive Recommendation. InProceedings of the 49th International ACM SIGIR Confer- ence on Research and Development in Information Retrieval. Association for Com- puting Machinery, New York, NY, USA, 369...

  6. [6]

    Shijie Geng, Shuchang Liu, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2022. Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt and Predict Paradigm (P5). InProceedings of the 16th ACM Conference on Recommender Systems. Association for Computing Machinery, New York, NY, USA, 299–315. doi:10.1145/3523227.3546767

  7. [7]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang

  8. [8]

    Yinhan He, Liam Collins, Bhuvesh Kumar, Jundong Li, Neil Shah, and Donald Loveland. 2026. Implicit Reasoning for Large Language Model-based Generative Recommendation. arXiv:2606.14142 [cs.CL] https://arxiv.org/abs/2606.14142

Show all 47 references
  1. [9]

    Yingzhi He, Yan Sun, Junfei Tan, Yuxin Chen, Xiaoyu Kong, Chunxu Shen, Xi- ang Wang, An Zhang, and Tat-Seng Chua. 2026. Reasoning over Semantic IDs Enhances Generative Recommendation. arXiv:2603.23183 [cs.IR] https: //arxiv.org/abs/2603.23183 Accepted by the 32nd ACM SIGKDD Co...

  2. [10]

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

  3. [11]

    Hervé Jégou, Matthijs Douze, and Cordelia Schmid. 2011. Product Quantization for Nearest Neighbor Search.IEEE Transactions on Pattern Analysis and Machine Intelligence33, 1 (2011), 117–128. doi:10.1109/TPAMI.2010.57

  4. [12]

    2011.Thinking, Fast and Slow

    Daniel Kahneman. 2011.Thinking, Fast and Slow. Farrar, Straus and Giroux, New York, NY, USA

  5. [13]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-Attentive Sequential Recom- mendation. In2018 IEEE International Conference on Data Mining. IEEE, Piscat- away, NJ, USA, 197–206. doi:10.1109/ICDM.2018.00035

  6. [14]

    Sunkyung Lee, Minjin Choi, Eunseong Choi, Hye-young Kim, and Jongwuk Lee

  7. [15]

    Zhanyu Liu, Shiyao Wang, Xingmei Wang, Rongzhou Zhang, Jiaxin Deng, Honghui Bao, Jinghao Zhang, Wuchao Li, PengFei Zheng, Xiangyu Wu, Yifei Hu, Qigen Hu, Xinchen Luo, Lejian Ren, Zhang Zixing, Qianqian Wang, Kuo Cai, Yunfan Wu, Hongtao Cheng, Zexuan Cheng, Lu Ren, Huanjie Wang...

  8. [16]

    Chen Ma, Peng Kang, and Xue Liu. 2019. Hierarchical Gating Networks for Se- quential Recommendation. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, New York, NY, USA, 825–833. doi:10.11...

  9. [17]

    Yu. A. Malkov and D. A. Yashunin. 2020. Efficient and Robust Approximate Nearest Neighbor Search Using Hierarchical Navigable Small World Graphs. IEEE Transactions on Pattern Analysis and Machine Intelligence42, 4 (2020), 824–

  10. [18]

    Jiabao Pan, Yan Zhang, Chen Zhang, Zuozhu Liu, Hongwei Wang, and Haizhou Li. 2024. DynaThink: Fast or Slow? A Dynamic Decision-Making Framework for Large Language Models. InProceedings of the 2024 Conference on Empirical Meth- ods in Natural Language Processing. Association fo...

  11. [19]

    Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan H. Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Maheswaran Sathiamoorthy. 2023. Recommender Systems with Generative Retrieval. InAdvances in Neural Informa...

  12. [20]

    Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme

  13. [21]

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

  14. [22]

    Jiakai Tang, Sunhao Dai, Teng Shi, Jun Xu, Xu Chen, Wen Chen, Jian Wu, and Yuning Jiang. 2026. Think Before Recommend: Unleashing the Latent Reasoning Power for Sequential Recommendation. IEEE Transactions on Knowledge and Data Engineering, Early Access. doi:10.1109/TKDE.2026.3694421

  15. [23]

    Tran, Mostafa Dehghani, Jianmo Ni, Dara Bahri, Harsh Mehta, Zhen Qin, Kai Hui, Zhe Zhao, Jai Gupta, Tal Schuster, William W

    Yi Tay, Vinh Q. Tran, Mostafa Dehghani, Jianmo Ni, Dara Bahri, Harsh Mehta, Zhen Qin, Kai Hui, Zhe Zhao, Jai Gupta, Tal Schuster, William W. Cohen, and Donald Metzler. 2022. Transformer Memory as a Differentiable Search Index. In Advances in Neural Information Processing Syste...

  16. [24]

    Aaron van den Oord, Oriol Vinyals, and Koray Kavukcuoglu. 2017. Neural Discrete Representation Learning. InAdvances in Neural Information Processing Systems, Vol. 30. Curran Associates, Inc., Red Hook, NY, USA, 6306–6315

  17. [25]

    Wenjie Wang, Honghui Bao, Xinyu Lin, Jizhi Zhang, Yongqi Li, Fuli Feng, See- Kiong Ng, and Tat-Seng Chua. 2024. Learnable Item Tokenization for Generative Recommendation. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management. Associatio...

  18. [26]

    Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019. Neural Graph Collaborative Filtering. InProceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval. Association for Computing Machinery, New York, NY, U...

  19. [27]

    Le, Ed H

    Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V. Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. Self-Consistency Improves Chain of Thought Reasoning in Language Models. The Eleventh International Conference on Learning Representations. https://openrev...

  20. [28]

    Ye Wang, Jiahao Xun, Minjie Hong, Jieming Zhu, Tao Jin, Wang Lin, Haoyuan Li, Linjun Li, Yan Xia, Zhou Zhao, and Zhenhua Dong. 2024. EAGER: Two- Stream Generative Recommender with Behavior-Semantic Collaboration. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Di...

  21. [29]

    Chi, Quoc V

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, and Denny Zhou. 2022. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models. InAdvances in Neural Information Processing Systems, Vol. 35. Curran Associates, I...

  22. [30]

    Wei Wei, Xubin Ren, Jiabin Tang, Qinyong Wang, Lixin Su, Suqi Cheng, Junfeng Wang, Dawei Yin, and Chao Huang. 2024. LLMRec: Large Language Models with Graph Augmentation for Recommendation. InProceedings of the 17th ACM Inter- national Conference on Web Search and Data Mining....

  23. [31]

    Likang Wu, Zhi Zheng, Zhaopeng Qiu, Hao Wang, Hongchao Gu, Tingjia Shen, Chuan Qin, Chen Zhu, Hengshu Zhu, Qi Liu, Hui Xiong, and Enhong Chen. 2024. A Survey on Large Language Models for Recommendation.World Wide Web27, 5, Article 60 (2024), 31 pages. doi:10.1007/s11280-024-01291-2

  24. [32]

    Shitao Xiao, Zheng Liu, Peitian Zhang, Niklas Muennighoff, Defu Lian, and Jian-Yun Nie. 2024. C-Pack: Packed Resources for General Chinese Embeddings. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. Association...

  25. [33]

    Jiawen Xie, Haiyang Wu, Deyi Ji, Yuekui Yang, and Shaoping Ma. 2025. LOHRec: Leveraging Order and Hierarchy in Generative Sequential Recommendation. In Findings of the Association for Computational Linguistics: EMNLP 2025. Association for Computational Linguistics, Suzhou, Chi...

  26. [34]

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

  27. [35]

    Griffiths, Yuan Cao, and Karthik Narasimhan

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. 2023. Tree of Thoughts: Deliberate Prob- lem Solving with Large Language Models. InAdvances in Neural Informa- tion Processing Systems, Vol. 36. Curran Associates, Inc., Re...

  28. [36]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. The Eleventh International Conference on Learning Representations. https://openreview.net/forum?id=WE_vluYUL-X Publishe...

  29. [37]

    Jiaqi Zhai, Lucy Liao, Xing Liu, Yueming Wang, Rui Li, Xuan Cao, Leon Gao, Zhaojie Gong, Fangda Gu, Jiayuan He, Yinghai Lu, and Yu Shi. 2024. Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations. InProceedings of the 41st In...

  30. [38]

    Yingyi Zhang, Junyi Li, Yejing Wang, Wenlin Zhang, Xiaowei Qian, Sheng Zhang, Yue Feng, Yichao Wang, Yong Liu, Xiangyu Zhao, and Xianneng Li. 2026. RAGR: Review-Augmented Generative Recommendation. arXiv:2605.17267 [cs.IR] https: //arxiv.org/abs/2605.17267

  31. [39]

    Le, and Ed H

    Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Claire Cui, Olivier Bousquet, Quoc V. Le, and Ed H. Chi

  32. [836]

    doi:10.1109/TPAMI.2018.2889473

  33. [2009]

    InProceed- ings of the 25th Conference on Uncertainty in Artificial Intelligence

    BPR: Bayesian Personalized Ranking from Implicit Feedback. InProceed- ings of the 25th Conference on Uncertainty in Artificial Intelligence. AUAI Press, Arlington, Virginia, USA, 452–461

  34. [2016]

    In- ternational Conference on Learning Representations

    Session-based Recommendations with Recurrent Neural Networks. In- ternational Conference on Learning Representations. https://openreview.net/ forum?id=yoffK5KZSgQ Published as a conference paper at ICLR 2016

  35. [2019]

    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. Association for Comput- ing Machinery, New York, NY, USA, 1441–1450. doi:10.1145/3...

  36. [2020]

    InProceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval

    LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation. InProceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. Association for Computing Machinery, New York, NY, USA, 639–648. doi:10.1145/3397...

  37. [2023]

    The Eleventh International Conference on Learning Representations

    Least-to-Most Prompting Enables Complex Reasoning in Large Language Models. The Eleventh International Conference on Learning Representations. https://openreview.net/forum?id=WZH7099tgfM Published as a conference paper at ICLR 2023. Received 20 February 2007; revised 12 March ...

  38. [2025]

    InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

    GRAM: Generative Recommendation via Semantic-aware Multi-granular Late Fusion. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Association for Computational Linguistics, Vienna, Austria, 33294–33312. doi:10.186...

  39. [3254]

    doi:10.1145/3637528.3671775

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.