Pith. sign in

REVIEW 3 major objections 6 minor 37 references

Semantic Convergence: Harmonizing Recommender Systems via Two-Stage Alignment and Behavioral Semantic Tokenization

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read This paper claims that a two-stage alignment framework, combining cascaded codebook tokenization with alignment fine-tuning and negative sampling, makes an LLM-based recommender outperform the previous best LLM-based method, LC-Rec, on…

desk verdict A clean incremental follow-up to LC-Rec with a plausible but under-supported headline gain; the evaluation protocol and hyperparameter selection need to be pinned down before the numbers can be trusted. read the letter →

arxiv 2412.13771 v1 pith:U6U3SRIG submitted 2024-12-18 cs.IR cs.AIcs.CL

classification cs.IRcs.AIcs.CL
keywords largelanguagemodelsrecommendersystemsalignmenttokenizationcascadedcodebooksnegativesamplingsequentialrecommendationsemantic
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

This paper proposes a two-stage framework that lets large language models (LLMs) act directly as recommenders without being overwhelmed by large item catalogs. The first stage, Alignment Tokenization, compresses item IDs into short sequences of shared, discrete codes using cascaded codebooks, and aligns those codes with the LLM's input embedding space. The second stage, Alignment Tasks, fine-tunes the LLM on prompts that mix user behavior, item text, and queries, including a negative-sampling task that imitates the generalization behavior of traditional recommenders. The authors claim the resulting system beats the previous best LLM-based recommender, LC-Rec, on all three Amazon datasets (Games, Arts, Instruments), with relative HR@1 gains from 3.25% to 15.77% and similar NDCG improvements. The framework also pre-caches per-user top-K item codes at inference time, making online deployment practical.

What carries the argument

The load-bearing object is the Alignment Tokenization module: a cascaded set of four codebooks, each with 256 codes, that maps an item's behavioral embedding (from a contrastive collaborative filtering model) into a four-token index sequence via residual quantization. The first codebook captures the coarsest semantics and each subsequent codebook approximates the residual, so related items share codes. A second, LLM alignment loss penalizes the distance between each chosen code and the mean-pooled LLM embedding of the item's title and description, pulling the code space toward the LLM's semantic space. The other half of the mechanism is the Alignment Task suite: sequential, text, and query alignment prompts plus a negative sampling task, all used to fine-tune the LLM with cross-entropy loss. Together these turn LLM fine-tuning into a tractable ranking problem over a compact, shared vocabulary.

What would settle it

An independent re-run of the Games experiment with full-corpus ranking (scoring all 16,859 items, not only items whose codes appear in the beam-search output) and with the codebook count and negative-sampling ratio chosen on a validation split would settle whether the reported gains over LC-Rec are genuine.

Watch

Extended reading notes

Core claim

The central claim is that the gap between sparse collaborative semantics and dense LLM token representations can be closed by a two-stage alignment. In the paper's telling, representing each item as four codes drawn from four cascaded codebooks (256 codes per book) makes the LLM vocabulary compact enough to fine-tune, while an LLM alignment loss during codebook training pre-synchronizes the code space with the LLM's embeddings. A set of supervised alignment tasks, namely sequential prediction, text-to-item translation, query-to-item matching, and a deliberately added negative-behavior task, then teach the LLM to combine behavioral and textual signals. The paper reports that this approach yields top Hit Ratio and NDCG on three Amazon review datasets, improving on LC-Rec by 3.25% to 15.77% depending on metric and dataset. The same tokenization also supports practical inference by pre-caching beam-searched top-K codes per user.

Load-bearing premise

The reported gains rest on the evaluation being fair and transferable: the paper does not state how HR/NDCG are computed over the item corpus (full ranking vs. ranking within a candidate set), and the negative-sampling ratio and number of codebooks are tuned on the Games dataset that later appears as the headline result.

Editorial extensions

If this is right

  • LLM-based recommenders can handle industrial-scale item pools because shared codes keep new-token counts small: 16,859 items fit in 1,024 new tokens with zero collisions.
  • Adding an LLM alignment loss during tokenization improves downstream recommendation accuracy, since the code space starts closer to the LLM's representation space.
  • Negative sampling during LLM fine-tuning adds training signal and mitigates sample selection bias, as shown by the ablation that removes it.
  • Larger LLMs scale recommendation quality: Llama-13B outperforms Llama-7B, suggesting LLM-based recommenders inherit a scaling law.
  • Pre-caching top-K beam-search codes per user makes online inference fast, and new items can be added by tokenizing them without retraining the LLM or codebooks.

Reading between the lines

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

  • The reported superiority over LC-Rec may partly reflect a comparison set: if the metrics rank only items whose code sequences appear in the beam-search output, traditional baselines that score the full corpus are at a disadvantage.
  • The token-sharing property of the codebooks suggests a cross-domain transfer test the paper does not run: items from an unseen domain could be tokenized with the same codebooks and fed to the fine-tuned LLM to probe whether the alignment generalizes.
  • The ablation tunes negative-sampling ratio and codebook count on Games and then reports Games as the main result; validating those choices on a held-out split would give a stricter estimate of the gains.
  • Because the inference cache stores per-user top-K codes, a user's evolving interests between cache refreshes are not reflected; the paper notes this as a storage and latency constraint but leaves hybrid strategies (cache plus vector retrieval for new items) untested.
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

3 major / 6 minor

Summary. The paper proposes a two-stage framework for LLM-based recommendation. In the first stage, Alignment Tokenization maps item embeddings (from DCCF) into a compact sequence of discrete codes drawn from cascaded codebooks, with an additional loss aligning the code embeddings with the LLM's input embedding space. In the second stage, Alignment Task fine-tunes Llama-7B on several instruction-style tasks (sequential, text, query) plus a negative-sampling task, using the learned code indices as item tokens. For inference, the paper pre-caches the top-K beam-search outputs per user to reduce online latency. The main empirical claim is that the method outperforms LC-Rec and other baselines on Amazon Games, Arts, and Instruments in HR@K and NDCG@K, with relative gains over LC-Rec ranging from +3.25% to +15.77%.

Significance. The work addresses a genuine problem in LLM-based recommendation, namely representing large item vocabularies as discrete tokens that are learnable by an LLM, and it does so with a clear, well-motivated design. The ablation studies (Table 3) and the 'behavior-only' control (Table 8) are useful and help isolate the contributions of the alignment loss and negative sampling. The pre-caching inference strategy is a practical step toward deployment. However, the empirical claims are not yet supported by the evidence as presented: the evaluation protocol is underspecified, the hyperparameters appear to be tuned on the same Games dataset used for the headline result, and no statistical significance is reported. If these issues are resolved, this could be a solid contribution; as it stands, the central comparison is not fully credible.

major comments (3)
  1. [Experiment Setup, Metrics] The paper never specifies how HR@K and NDCG@K are computed over the item corpus. It is essential to state whether the ground-truth item is ranked against the full item set, a sampled candidate set, or only the items produced by each method's own generation/beam-search procedure. Since LC-Rec and Ours decode item-code sequences via beam search while SASRec/BERT4Rec/DCCF score all items, the candidate sets can differ in size and composition. Without a common evaluation protocol, the numbers in Table 1 are not directly comparable across methods, and the headline claim of improvement over LC-Rec is under-supported. Please specify the protocol precisely and, if possible, provide evaluation code or a concrete example.
  2. [Ablation Study, Tables 4 and 5] The negative sampling ratio (1:3) and the number of codebooks (4) are selected by comparing test metrics on the Games dataset (Tables 4 and 5), and the same Games dataset is used as the first block of Table 1. No validation split is described. If these hyperparameters were chosen based on test-set performance, the reported Games improvements (e.g., +15.77% HR@1) are at least partly an artifact of test-set selection. Please clarify whether a validation split was used, or re-run the selection on a hold-out set and report the resulting performance for all datasets.
  3. [Experiment Results, Table 1] The reported differences between Ours and LC-Rec are small for several metrics (Instruments HR@1 +3.25%; Arts HR@5 +4.20%). The paper reports no error bars, standard deviations, or significance tests. It is therefore unclear whether the observed improvements are statistically meaningful or within run-to-run variance. Please report multiple independent runs with standard errors, or at least a paired significance test (e.g., bootstrap or t-test) for the main comparison in Table 1.
minor comments (6)
  1. [Method, Alignment Tokenization; Table 5] The notation is inconsistent: in the Method section and Eq. (3), N denotes the number of codebook levels and C denotes the number of codes per level, with N=4 and C=256. In Table 5, however, C denotes the number of codebooks. This makes Table 5 hard to interpret; please use disjoint symbols for the two quantities.
  2. [Ablation Study] The text states that 'our method shows a slight improvement compared to "w/o AM"', but Table 3 does not include a 'w/o AM' row; it includes 'ED', 'w/o LA', and 'w/o NS'. This appears to be a typo for 'w/o LA'. Please correct the text or the table.
  3. [Abstract and Ablation Study] There are several typos, including 'effciency' in the abstract and 'negtive sampling' in the Ablation Study section. Please proofread the manuscript.
  4. [Inference, cold-start discussion] The paragraph claiming that new items can be handled without retraining the LLM or codebooks ('mitigating the cold start problem') is not supported by any experiment. Please either add a small experiment demonstrating cold-start capability or soften the claim.
  5. [Experiment Setup] The paper does not report training hyperparameters for the LLM fine-tuning stage, such as LoRA rank, learning rate, batch size, number of epochs, or hardware used. Including these details would improve reproducibility.
  6. [Inference] The inference prompt is written as 'The user has interacted with <itemi>,... in chronological order . Can you predict the next possible item that the user may expect?' with an extra space before the period; the surrounding sentence also contains 'which mentioned in LC-Rec'. Please correct the grammar and punctuation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is an empirical comparison on external Amazon datasets against external baselines, and the derivation chain does not reduce to its inputs.

full rationale

The paper's derivation chain is self-contained in the sense required by the circularity pass. The Alignment Tokenization module builds cascaded codebooks by quantizing item embeddings from DCCF (Ren et al. 2023) with a residual distance loss L_B and an LLM-alignment loss L_L (Eqs. 1-4); these embeddings are external inputs used as features, not a disguised version of the target next-item prediction. The Alignment Task then fine-tunes Llama-7B with sequential, text, query, and negative-sampling prompts, and evaluation designates the user's last interacted item as the label with preceding actions as input, which is a standard held-out empirical protocol. No equation equates a reported HR/NDCG result to a fitted constant, and no parameter fitted to a subset is later renamed as a prediction; the negative-sampling ratio and codebook count in Tables 4 and 5 are hyperparameter choices for the implemented method, and any concern about test-set selection or an underspecified candidate-set metric is an experimental-validity issue rather than circularity in the derivation. All load-bearing technical citations (LC-Rec, DCCF, RQ-VAE, Llama) are to external work, and no uniqueness theorem or prior result from the present authors is invoked to forbid alternatives. The closest in-scope observation is that the codebooks are trained on the same behavior data later used for evaluation, but that is supervised feature construction and does not make the reported improvements true by definition.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the choice of codebook hyperparameters (N=4 codebooks, 256 codes per layer), the negative-sampling ratio (1:3, tuned on Games), and the domain assumption that DCCF behavior embeddings plus LLM text embeddings can be reconciled through shared code indices. It also rests on standard supervised-learning assumptions: that the training distribution resembles the evaluation distribution and that excluding long prompts does not bias the metrics. No new physical entities are introduced; the 'Alignment Tokenization' and 'Alignment Tasks' are algorithmic components, not independent postulates.

free parameters (8)
  • number_of_codebooks (N) = 4
    Chosen on Games data (Table 5) as the balance between collision rate (0.0% at N=4) and performance; N=4 used for all reported results.
  • codebook_size (codes per layer) = 256
    Set in experiments; each item is represented by 4 indices, one from each of 4 codebooks of 256 codes.
  • negative_sampling_ratio (positive:negative) = 1:3
    Tuned on Games dataset (Table 4); 1:3 chosen for diminishing returns though 1:4 gives marginally higher HR@5 (0.0871 vs 0.0858).
  • max_user_sequence_length = 20
    Preprocessing choice to mitigate long-tail sequences; long users (4-5% of data) are truncated.
  • prompt_length_threshold = 2048
    Samples with prompts exceeding 2048 tokens are excluded; exclusion count not reported.
  • DCCF_item_embedding_dim = 32
    Dimension of the behavior-based item embeddings that are quantized into codebooks.
  • alignment_loss_weight (L^L vs L^B) = 1 (implicit)
    The total loss in Eq. (4) is a simple sum of behavior and LLM alignment losses; no weighting scheme is described.
  • distance_function = cosine
    Used in Dist() in Eqs. (1)-(3); no comparison to other distances.
assumptions (6)
  • domain assumption DCCF item embeddings trained exclusively on behavior capture collaborative semantics that can be quantized into discrete codebooks.
    Invoked in Alignment Tokenization: 'we utilize embeddings derived from DCCF trained exclusively on behavior as the item embeddings.'
  • domain assumption Mean-pooled LLM input embeddings of item titles/descriptions are a valid semantic target for codebook alignment.
    Eq. (3) defines E^LLM_i as the average pooling of the LLM input layer embedding through titles and descriptions; no evidence that mean pooling is the right target.
  • domain assumption Fine-tuning Llama-7B with cross-entropy on next-item code prediction yields a good recommender.
    Used throughout Alignment Task section; no comparison to other fine-tuning objectives beyond the provided ablations.
  • domain assumption Excluding samples with prompt length > 2048 does not bias results.
    Stated in Experiment Setup without reporting the number of excluded samples.
  • standard math RQ-VAE residual quantization converges to a useful discrete approximation of the embedding space.
    Borrowed from Rajput et al. 2024; the paper cites it as the basis for the cascaded codebooks.
  • ad hoc to paper The LLM can generate meaningful item-code sequences for items added after fine-tuning (cold start).
    Asserted in the Inference section without experiments; new items share the same code tokens, but the LLM has not been trained on their specific code combinations, so this is a testable but untested assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Semantic Convergence: Harmonizing Recommender Systems via Two-Stage Alignment and Behavioral Semantic Tokenization." pith.science (2026). https://pith.science/paper/U6U3SRIG

@misc{pith2026241213771,
  author       = {Pith},
  title        = {Pith review of: Semantic Convergence: Harmonizing Recommender Systems via Two-Stage Alignment and Behavioral Semantic Tokenization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U6U3SRIG}},
  note         = {Machine review of arXiv:2412.13771}
}
read the original abstract

Large language models (LLMs), endowed with exceptional reasoning capabilities, are adept at discerning profound user interests from historical behaviors, thereby presenting a promising avenue for the advancement of recommendation systems. However, a notable discrepancy persists between the sparse collaborative semantics typically found in recommendation systems and the dense token representations within LLMs. In our study, we propose a novel framework that harmoniously merges traditional recommendation models with the prowess of LLMs. We initiate this integration by transforming ItemIDs into sequences that align semantically with the LLMs space, through the proposed Alignment Tokenization module. Additionally, we design a series of specialized supervised learning tasks aimed at aligning collaborative signals with the subtleties of natural language semantics. To ensure practical applicability, we optimize online inference by pre-caching the top-K results for each user, reducing latency and improving effciency. Extensive experimental evidence indicates that our model markedly improves recall metrics and displays remarkable scalability of recommendation systems.

Figures

Figures reproduced from arXiv: 2412.13771 by the authors.

Figure 1
Figure 1. Illustration of LLM-Based Recommendation [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The pipeline of ours two-stage alignment for recommendation. During the Alignment Tokenization phase, our objec [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Inference Stage. We cache valid codes in predic [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 16 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Dao, T.; Fu, D.; Ermon, S.; Rudra, A.; and R \'e , C. 2022. Flashattention: Fast and memory-efficient exact attention with io-awareness. Advances in Neural Information Processing Systems, 35: 16344--16359

  4. [4]

    Di Palma, D. 2023. Retrieval-augmented recommender system: Enhancing recommender systems with large language models. In Proceedings of the 17th ACM Conference on Recommender Systems, 1369--1373

  5. [5]

    Geng, S.; Liu, S.; Fu, Z.; Ge, Y.; and Zhang, Y. 2022. Recommendation as language processing (rlp): A unified pretrain, personalized prompt & predict paradigm (p5). In RecSys

  6. [6]

    He, X.; Deng, K.; Wang, X.; Li, Y.; Zhang, Y.; and Wang, M. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. In Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval, 639--648

  7. [7]

    Hidasi, B.; Karatzoglou, A.; Baltrunas, L.; and Tikk, D. 2015. Session-based recommendations with recurrent neural networks. arXiv preprint arXiv:1511.06939

  8. [8]

    Hou, Y.; He, Z.; McAuley, J.; and Zhao, W. X. 2023. Learning vector-quantized item representation for transferable sequential recommenders. In WWW

Show all 37 references
  1. [9]

    X.; Li, Y.; Ding, B.; and Wen, J.-R

    Hou, Y.; Mu, S.; Zhao, W. X.; Li, Y.; Ding, B.; and Wen, J.-R. 2022. Towards universal sequence representation learning for recommender systems. In SIGKDD

  2. [10]

    J.; Shen, Y.; Wallis, P.; Allen-Zhu, Z.; Li, Y.; Wang, S.; Wang, L.; and Chen, W

    Hu, E. J.; Shen, Y.; Wallis, P.; Allen-Zhu, Z.; Li, Y.; Wang, S.; Wang, L.; and Chen, W. 2021. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685

  3. [11]

    Hua, W.; Xu, S.; Ge, Y.; and Zhang, Y. 2023. How to Index Item IDs for Recommendation Foundation Models. SIGIR-AP

  4. [12]

    Kang, W.-C.; and McAuley, J. 2018. Self-attentive sequential recommendation. In 2018 IEEE international conference on data mining (ICDM), 197--206. IEEE

  5. [13]

    Li, X.; Chen, C.; Zhao, X.; Zhang, Y.; and Xing, C. 2023. E4SRec: An elegant effective efficient extensible solution of large language models for sequential recommendation. arXiv preprint arXiv:2312.02443

  6. [14]

    Lin, X.; Wang, W.; Li, Y.; Feng, F.; Ng, S.-K.; and Chua, T.-S. 2023. A Multi-facet Paradigm to Bridge Large Language Model and Recommendation. arXiv preprint arXiv:2310.06491

  7. [15]

    Luo, S.; Yao, Y.; He, B.; Huang, Y.; Zhou, A.; Zhang, X.; Xiao, Y.; Zhan, M.; and Song, L. 2024. Integrating Large Language Models into Recommendation via Mutual Augmentation and Adaptive Aggregation. arXiv preprint arXiv:2401.13870

  8. [16]

    Lyu, H.; Jiang, S.; Zeng, H.; Xia, Y.; and Luo, J. 2023. Llm-rec: Personalized recommendation via prompting large language models. arXiv preprint arXiv:2307.15780

  9. [17]

    Ma, X.; Zhao, L.; Huang, G.; Wang, Z.; Hu, Z.; Zhu, X.; and Gai, K. 2018. Entire space multi-task model: An effective approach for estimating post-click conversion rate. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, 1137--1140

  10. [18]

    Ni, J.; Li, J.; and McAuley, J. 2019. Justifying recommendations using distantly-labeled reviews and fine-grained aspects. In EMNLP-IJCNLP

  11. [19]

    Rajput, S.; Mehta, N.; Singh, A.; Hulikal Keshavan, R.; Vu, T.; Heldt, L.; Hong, L.; Tay, Y.; Tran, V.; Samost, J.; et al. 2024. Recommender systems with generative retrieval. Advances in Neural Information Processing Systems, 36

  12. [20]

    Ren, X.; Wei, W.; Xia, L.; Su, L.; Cheng, S.; Wang, J.; Yin, D.; and Huang, C. 2024. Representation learning with large language models for recommendation. In Proceedings of the ACM on Web Conference 2024, 3464--3475

  13. [21]

    Ren, X.; Xia, L.; Zhao, J.; Yin, D.; and Huang, C. 2023. Disentangled contrastive collaborative filtering. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval, 1137--1146

  14. [22]

    Sun, F.; Liu, J.; Wu, J.; Pei, C.; Lin, X.; Ou, W.; and Jiang, P. 2019. BERT4Rec: Sequential recommendation with bidirectional encoder representations from transformer. In CIKM

  15. [23]

    Touvron, H.; Lavril, T.; Izacard, G.; Martinet, X.; Lachaux, M.-A.; Lacroix, T.; Rozi \`e re, B.; Goyal, N.; Hambro, E.; Azhar, F.; et al. 2023. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971

  16. [24]

    Wang, H.; Liu, X.; Fan, W.; Zhao, X.; Kini, V.; Yadav, D.; Wang, F.; Wen, Z.; Tang, J.; and Liu, H. 2024 a . Rethinking Large Language Model Architectures for Sequential Recommendations. arXiv preprint arXiv:2402.09543

  17. [25]

    Wang, X.; He, X.; Wang, M.; Feng, F.; and Chua, T.-S. 2019. Neural graph collaborative filtering. In Proceedings of the 42nd international ACM SIGIR conference on Research and development in Information Retrieval, 165--174

  18. [26]

    Wang, X.; Wu, L.; Hong, L.; Liu, H.; and Fu, Y. 2024 b . LLM-Enhanced User-Item Interactions: Leveraging Edge Information for Optimized Recommendations. arXiv preprint arXiv:2402.09617

  19. [27]

    Wei, W.; Ren, X.; Tang, J.; Wang, Q.; Su, L.; Cheng, S.; Wang, J.; Yin, D.; and Huang, C. 2024. Llmrec: Large language models with graph augmentation for recommendation. In Proceedings of the 17th ACM International Conference on Web Search and Data Mining, 806--815

  20. [28]

    Wenyueh. 2023. LLM-RecSys-ID. https://github.com/Wenyueh/LLM-RecSys-ID/, Last accessed on 08-15-2024

  21. [29]

    J.; and Jing, H

    Wu, C.-Y.; Ahmed, A.; Beutel, A.; Smola, A. J.; and Jing, H. 2017. Recurrent Recommender Networks. In Proceedings of the Tenth ACM International Conference on Web Search and Data Mining, 495--503

  22. [30]

    Xi, Y.; Liu, W.; Lin, J.; Zhu, J.; Chen, B.; Tang, R.; Zhang, W.; Zhang, R.; and Yu, Y. 2023. Towards open-world recommendation with knowledge augmentation from large language models. arXiv preprint arXiv:2306.10933

  23. [31]

    Zhang, C.; Wu, S.; Zhang, H.; Xu, T.; Gao, Y.; Hu, Y.; and Chen, E. 2024 a . NoteLLM: A Retrievable Large Language Model for Note Recommendation. In Companion Proceedings of the ACM on Web Conference 2024, 170--179

  24. [32]

    Zhang, C.; Zhang, H.; Wu, S.; Wu, D.; Xu, T.; Gao, Y.; Hu, Y.; and Chen, E. 2024 b . NoteLLM-2: Multimodal Large Representation Models for Recommendation. arXiv preprint arXiv:2405.16789

  25. [33]

    Zhang, Y.; Feng, F.; Zhang, J.; Bao, K.; Wang, Q.; and He, X. 2023. CoLLM: Integrating Collaborative Embeddings into Large Language Models for Recommendation. arXiv preprint arXiv:2310.19488

  26. [34]

    X.; and Wen, J.-R

    Zheng, B.; Hou, Y.; Lu, H.; Chen, Y.; Zhao, W. X.; and Wen, J.-R. 2023. Adapting large language models by integrating collaborative semantics for recommendation. arXiv preprint arXiv:2311.09049

  27. [35]

    X.; Zhu, Y.; Wang, S.; Zhang, F.; Wang, Z.; and Wen, J.-R

    Zhou, K.; Wang, H.; Zhao, W. X.; Zhu, Y.; Wang, S.; Zhang, F.; Wang, Z.; and Wen, J.-R. 2020. S3-rec: Self-supervised learning for sequential recommendation with mutual information maximization. In CIKM

  28. [36]

    X.; and Wen, J.-R

    Zhou, K.; Yu, H.; Zhao, W. X.; and Wen, J.-R. 2022. Filter-enhanced MLP is all you need for sequential recommendation. In WWW

  29. [37]

    Zhu, Y.; Wu, L.; Guo, Q.; Hong, L.; and Li, J. 2024. Collaborative large language model for recommender systems. In Proceedings of the ACM on Web Conference 2024, 3162--3172

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.