Pith. sign in

REVIEW 4 major objections 6 minor 64 references

ABXI: Invariant Interest Adaptation for Task-Guided Cross-Domain Sequential Recommendation

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read ABXI claims that aligning sequences to the target item's domain, not to timestamps, and extracting domain-invariant interests with LoRA adapters, significantly improves dual-target cross-domain sequential recommendation.

desk verdict Task-guided alignment is a genuine fix, but the reported gains conflate alignment with architecture; needs aligned baselines before claiming SOTA. read the letter →

arxiv 2501.15118 v2 pith:P5AGRLIS submitted 2025-01-25 cs.IR

Qingtian Bian , Marcus Vinícius de Carvalho , Tieying Li , Jiaxing Xu , Hui Fang , Yiping Ke This is my paper · ORCID
classification cs.IR
keywords cross-domainsequentialrecommendationdual-targetdomain-invariantinterestlow-rankadaptationtask-guidedalignmentpredictionmismatchself-attentionencoder
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

The paper tries to establish that the standard way of aligning cross-domain and domain-specific sequences in cross-domain sequential recommendation—by timestamp—causes prediction mismatches that hurt accuracy, and that aligning instead to the domain of the target item, together with extracting and adapting only domain-invariant interests, removes the mismatch and yields large gains. On three Amazon datasets, the proposed model ABXI outperforms all baselines, including state-of-the-art cross-domain models, on HR, NDCG, and MRR with p<0.01. The authors' core message is that a single shared self-attention encoder with lightweight low-rank adapters is enough to capture both shared and domain-specific knowledge, if the sequences are properly aligned.

What carries the argument

The central mechanism is task-guided alignment: in the seq2seq split, the cross-domain sequence removes the last token of the raw sequence, and each domain-specific sequence is obtained by masking out the other domain's tokens and padding so that its positions line up with the same ground-truth tokens as the cross-domain sequence. This removes prediction mismatches where an input token from domain A is used to predict a target from domain B. Supporting machinery: two LoRA types—dLoRA adapters that switch a single shared encoder among domains X, A, and B, and iLoRA adapters inside an invariant projector that adapt extracted domain-invariant interests to each domain's recommendation head.

What would settle it

Design an experiment on a dataset with many clean cross-domain transitions, such as a movie watch immediately followed by a book purchase, and compare ABXI's task-guided alignment with a timestamp-aligned model that equally separates domain-specific and invariant encoders; if the timestamp-aligned model matches or beats ABXI on such data, the claim that cross-domain tokens only hurt prediction would be refuted.

Watch

Extended reading notes

Core claim

ABXI introduces task-guided alignment: instead of pairing each position in a domain-specific sequence with the same-timestamp position in the cross-domain sequence, it removes the last token to form the cross-domain sequence and derives each domain-specific sequence by masking the other domain's tokens and re-aligning position-wise to the same ground-truth tokens, so every input token's encoded domain-specific knowledge matches the domain of the token being predicted. This unifies the ground truths of cross-domain and domain-specific sequences into a single optimization objective. To transfer knowledge, a shared self-attention encoder is switched between domains with three domain LoRAs, and a separate invariant projector with two invariant LoRAs extracts domain-invariant interests from the cross-domain representation and adapts them to each domain's final recommendation. The paper claims this design is validated by large improvements over all baselines on the Food-Kitchen, Beauty-Electronics, and Movie-Book datasets.

Load-bearing premise

The load-bearing premise is that a token from the other domain always carries harmful domain-specific knowledge, so masking it out of the domain-specific sequence and aligning each sequence to its own target domain strictly improves predictions rather than discarding useful information.

Editorial extensions

If this is right

  • Task-guided alignment unifies cross-domain and domain-specific losses into one InfoNCE objective, simplifying the training setup for dual-target cross-domain sequential recommendation.
  • A single shared self-attention encoder with domain LoRAs suffices to model both cross-domain and domain-specific sequences, removing the need for separate per-domain encoders.
  • Domain-invariant interest extraction with invariant LoRA adapters transfers knowledge without carrying domain-specific noise, which should be especially helpful when one target domain is sparse.
  • The approach generalizes beyond two domains: one shared encoder, N domain LoRAs, N invariant LoRAs, one invariant projector, and N domain-specific projectors.
  • Reverting to timestamp-guided alignment produces the largest performance drop among all ablations, directly supporting the paper's prediction-mismatch diagnosis.

Reading between the lines

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

  • Editorial extension: if the prediction-mismatch diagnosis is right, any architecture that fuses cross-domain and domain-specific representations at the token level may suffer the same issue, so task-guided alignment could serve as a drop-in fix in other multi-domain transformer models.
  • Editorial extension: the success of LoRA in single-stage (non-fine-tuning) training here suggests low-rank adapters are a general lightweight mechanism for multi-domain recommendation, worth testing in large-scale industrial settings with many domains.
  • Editorial extension: the paper does not test the setting where one domain is much sparser than the other; varying the sparsity ratio would provide a direct test of the predicted benefit of invariant interest transfer, since the authors' logic implies ABXI's margin should grow as the target domain gets sparser.
  • Editorial extension: the rank analysis shows optimal LoRA rank varies with dataset size, with smaller datasets favoring larger ranks, which suggests a testable hypothesis that under heavier data sparsity the low-rank assumption weakens and needs more capacity.
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 / 6 minor

Summary. The paper proposes ABXI, a dual-target cross-domain sequential recommender that combines task-guided sequence alignment with two LoRA-based adaptation mechanisms. Domain-specific sequences are remasked and realigned so that their position-wise ground truths match the cross-domain ground truth, avoiding what the authors call prediction mismatches in timestamp-aligned baselines. A shared self-attention encoder with per-domain LoRA modules encodes the cross-domain and domain-specific sequences, and a shared 'invariant' projector with two invariant LoRAs adapts cross-domain knowledge into each domain. The model is trained with a single InfoNCE loss. Experiments on three Amazon-derived dataset pairs (FK, BE, MB) compare ABXI with single- and dual-target SDSR and CDSR baselines, with ablations that remove or replace the alignment, dLoRAs, iLoRAs, and projectors.

Significance. If the claims hold, the paper makes two useful contributions: a simple preprocessing fix for the timestamp-alignment problem in dual-target CDSR, and evidence that LoRA-style adaptation can work in single-stage recommender training rather than only in LLM fine-tuning. The empirical protocol is largely standard: leave-one-out evaluation, five seeds, sampled negative items, and a released code repository. The ablations are informative and broadly support the importance of task-guided alignment and the LoRA modules. However, two central claims are currently stronger than the experiments warrant: the headline superiority over baselines is confounded with the new alignment preprocessing, and the word 'invariant' is used without a formal invariance constraint or measurement. Those gaps make the significance conditional until the controlled comparisons and definitions are supplied.

major comments (4)
  1. [§4.4, Table 3, and §3.3] The headline performance comparison conflates the task-guided alignment preprocessing with the ABXI architecture. All baselines are trained with their original timestamp-guided or standard sequence construction, while only ABXI uses the aligned sequences defined in Section 3.3 and the unified InfoNCE loss in Eq. (8). The Vts ablation in Table 4 shows that alignment matters within ABXI, but it does not answer whether a baseline such as SASRec-2 or DREAM would close most of the gap if it were trained on the same aligned sequences and the same unified loss. Please add this controlled comparison, or clearly reframe the main claim as 'ABXI with task-guided alignment outperforms baselines without it.' As written, the reported gains cannot be attributed specifically to the proposed model architecture.
  2. [§3.6, Eqs. (4)–(6), and title] The model is called 'invariant' and the paper claims that Proji 'extracts domain-invariant interests from cross-domain representations,' but no invariance constraint, domain-adversarial objective, or invariance metric is defined. Proji is simply a projector applied to the cross-domain encoding and shared by both domains; nothing prevents domain-specific information from flowing through it. Since 'invariant' is load-bearing in the title and in the motivation, please either add an operational invariance test (for example, train a domain classifier on the extracted representation and report its accuracy) or add an explicit invariance loss, or soften the terminology to 'shared cross-domain projector' throughout.
  3. [Table 4, rows Ve3 and Vdp3] The Ve3 and Vdp3 rows report identical means and standard deviations for the Beauty, Electronics, Movie, and Book domains (0.1729/0.1112/0.2114/0.1488 with the same uncertainties). These two variants are architecturally different—one replaces the shared encoder and dLoRAs with three encoders, the other replaces the dLoRAs with three projectors—so identical performance across four domains is difficult to explain unless the rows were copied. Please verify these numbers, because RQ3's conclusion that LoRA modules are preferable to dense layers relies on the correctness of the Vdp3 row.
  4. [§4.4, Table 3 caption] The paper states that all pairwise differences between ABXI and the best baseline are significant at p<0.01 using paired t-tests, but no details are given about what is paired, how many observations enter each test, or whether multiple-comparison corrections are applied across the many metrics, domains, and dataset pairs. With only five random seeds, a paired t-test has very low power and is sensitive to the seed set. Please report the test setup explicitly and either use more seeds with non-parametric tests or restrict the significance claim to the specific comparisons that survive a multiple-testing correction.
minor comments (6)
  1. [§4.5] The text refers to 'V5' as the timestamp-alignment ablation, but the listed variants are Vts, V1, V2, V3, and V4; please align the naming to avoid confusion.
  2. [§4.5 heading] The subsection title says 'LoRA versus Dense Layers (RQ4)', but this corresponds to RQ3; RQ4 is the rank analysis in Section 4.7.
  3. [Table 3, Movie domain] For SASRec-1 in the Movie domain, NDCG@10 is reported as 0.1647 while MRR is 0.1874; under the standard definitions NDCG@10 should be at least as large as MRR for binary relevance, so one of these numbers appears to be a typo.
  4. [Table 1] The Movie domain interaction count is printed as '347,65', which appears to be missing a digit; please correct it.
  5. [§3.3] The sentence 'we first extract the last interaction and the first interaction to form the cross-domain sequence seqX and the cross-domain ground truth gtX' is confusing as written; a standard seq2seq split removes the last item for the ground truth and keeps the preceding prefix as the sequence. Please rephrase to describe the split unambiguously.
  6. [§4.6 and Table 4] The variant names 'Vip3', 'Vi3', and 'Vip2' are used inconsistently (the text also writes 'V i3' and 'Vi2'); please settle on one naming convention and define each variant once in one place.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ABXI's reported gains are empirical comparisons on held-out data, and no prediction or fitted quantity is defined in terms of the method's own outputs.

full rationale

Walking the derivation chain, ABXI's components are defined independently of the evaluation metrics. Task-guided alignment is a sequence-splitting and masking rule (Section 3.3), while the dLoRA/iLoRA and projector modules are parameterized transformations (Sections 3.5-3.6). The optimization objective (Eq. 7-8) is an InfoNCE loss over held-out positive/negative samples; the paper explicitly states that gtA and gtB are only intermediate terms for realignment and are not used during optimization, so the unified ground truths are not fitted targets. LoRA ranks are tuned on validation MRR, but the headline claims are measured on test ground truths under leave-one-out evaluation, so there is no fitted-input-called-prediction pattern. The only self-citation, CPMR [7], appears in a related-work enumeration of sequential recommendation methods and is not load-bearing; the LoRA foundation [15] and SASRec backbone [20] are external results. The Vts ablation is an internal comparison against the proposed alignment, not a definitional equivalence. The skeptical concern that baseline comparisons do not isolate alignment preprocessing from architecture is an experimental-design and attribution issue, not a circularity of the paper's derivation. Therefore no specific circular step can be exhibited, and the correct finding is no significant circularity.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The model depends on standard deep learning components and a domain assumption about cross-domain token harmfulness. No new physical or mathematical entities are introduced. The free parameters are standard hyperparameters, with LoRA ranks being the most load-bearing since they are tuned per dataset.

free parameters (6)
  • LoRA rank r_d for dLoRA = 64 on FK, 16 on BE, 4 on MB
    Selected per dataset by grid search over {0,4,8,16,32,64,128,proj} as shown in Figure 4.
  • LoRA rank r_i for iLoRA = 64 on FK, 64 on BE, 4 on MB
    Selected per dataset by grid search over {0,4,8,16,32,64,128,proj} as shown in Figure 4.
  • Temperature tau = 0.75
    Set in Table 2; controls sharpness of InfoNCE loss.
  • Negative sample count N_neg = 128
    Set in Table 2; number of randomly sampled negative items per positive sample.
  • Embedding dimension d = 256
    Set in Table 2; dimension of item and position embeddings.
  • Encoder layers = 1
    Set in Table 2; number of self-attention layers in the shared encoder.
assumptions (4)
  • standard math Standard transformer self-attention and InfoNCE loss work as described in prior work
    Used without proof, citing Vaswani et al. and van den Oord et al.
  • domain assumption Users have overlapping interactions in both domains and their interests can be shared
    The problem formulation in Section 3.1 assumes overlapping users and dual-target prediction.
  • domain assumption A token from one domain always carries domain-specific knowledge that harms prediction of a target in the other domain
    This is the core premise for task-guided alignment (Section 3.3, Figure 2c), supported only by an ablation on three datasets.
  • ad hoc to paper The shared projector with iLoRA extracts domain-invariant interests
    Section 3.6, Eq. 5 defines the projector but provides no invariance constraint or formal definition; the name is aspirational.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ABXI: Invariant Interest Adaptation for Task-Guided Cross-Domain Sequential Recommendation." pith.science (2026). https://pith.science/paper/P5AGRLIS

@misc{pith2026250115118,
  author       = {Pith},
  title        = {Pith review of: ABXI: Invariant Interest Adaptation for Task-Guided Cross-Domain Sequential Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P5AGRLIS}},
  note         = {Machine review of arXiv:2501.15118}
}
read the original abstract

Cross-Domain Sequential Recommendation (CDSR) has recently gained attention for countering data sparsity by transferring knowledge across domains. A common approach merges domain-specific sequences into cross-domain sequences, serving as bridges to connect domains. One key challenge is to correctly extract the shared knowledge among these sequences and appropriately transfer it. Most existing works directly transfer unfiltered cross-domain knowledge rather than extracting domain-invariant components and adaptively integrating them into domain-specific modelings. Another challenge lies in aligning the domain-specific and cross-domain sequences. Existing methods align these sequences based on timestamps, but this approach can cause prediction mismatches when the current tokens and their targets belong to different domains. In such cases, the domain-specific knowledge carried by the current tokens may degrade performance. To address these challenges, we propose the A-B-Cross-to-Invariant Learning Recommender (ABXI). Specifically, leveraging LoRA's effectiveness for efficient adaptation, ABXI incorporates two types of LoRAs to facilitate knowledge adaptation. First, all sequences are processed through a shared encoder that employs a domain LoRA for each sequence, thereby preserving unique domain characteristics. Next, we introduce an invariant projector that extracts domain-invariant interests from cross-domain representations, utilizing an invariant LoRA to adapt these interests into modeling each specific domain. Besides, to avoid prediction mismatches, all domain-specific sequences are aligned to match the domains of the cross-domain ground truths. Experimental results on three datasets demonstrate that our approach outperforms other CDSR counterparts by a large margin. The codes are available in https://github.com/DiMarzioBian/ABXI.

Figures

Figures reproduced from arXiv: 2501.15118 by the authors.

Figure 1
Figure 1. Our proposal on generating recommendations by integrating domain-specific interests with domain-invariant interests [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the sequence splits under different alignments, where [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Proposed ABXI model. Recently, Large Language Models (LLMs) have attracted much attention for their strong performance and scalability [2, 44]. Re￾searchers have attempted to introduce LLMs into recommender systems as well [1, 28]. However, these LLMs are Pretrained Lan￾guage Models (PLMs) that are pretrained on Natural Language Processing (NLP) tasks. Therefore, researchers need to adapt these models to the recomme… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: MRR performance for varying rank values 𝑟𝑑 for dLoRA and 𝑟𝑖 for iLoRA (RQ4). the following four variants: Ve3: replace the shared encoder and dLoRAs with three self-attention encoders for domains 𝑋, 𝐴 and 𝐵, respectively; Vdp3: replace three dLoRAs with three projector…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 37 canonical work pages

  1. [1]

    Arkadeep Acharya, Brijraj Singh, and Naoyuki Onoe. 2023. Llm based generation of item-description for recommendation system. In Proceedings of the 17th ACM Conference on Recommender Systems . 1204–1207

    work page 2023

  2. [2]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Floren- cia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774 (2023)

    arXiv 2023

  3. [3]

    Armen Aghajanyan, Luke Zettlemoyer, and Sonal Gupta. 2020. Intrinsic dimen- sionality explains the effectiveness of language model fine-tuning. arXiv preprint arXiv:2012.13255 (2020)

    arXiv 2020

  4. [4]

    Nawaf Alharbi and Doina Caragea. 2021. Cross-domain attentive sequential recommendations based on general and current user preferences (cd-asr). In IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology. 48–55

    work page 2021

  5. [5]

    Nawaf Alharbi and Doina Caragea. 2022. Cross-domain self-attentive sequential recommendations. In Proceedings of International Conference on Data Science and Applications: ICDSA 2021, Volume 2. Springer, 601–614

    work page 2022

  6. [6]

    Keqin Bao, Jizhi Zhang, Yang Zhang, Wenjie Wang, Fuli Feng, and Xiangnan He. 2023. Tallrec: An effective and efficient tuning framework to align large language model with recommendation. InProceedings of the 17th ACM Conference on Recommender Systems. 1007–1014

    2023

  7. [7]

    Qingtian Bian, Jiaxing Xu, Hui Fang, and Yiping Ke. 2023. CPMR: Context-Aware Incremental Sequential Recommendation with Pseudo-Multi-Task Learning. In Proceedings of the 32nd ACM International Conference on Information and Knowl- edge Management. 120–130

    work page 2023

  8. [8]

    Jiangxia Cao, Xin Cong, Jiawei Sheng, Tingwen Liu, and Bin Wang. 2022. Con- trastive Cross-Domain Sequential Recommendation. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management. 138–147

    work page 2022

Show all 64 references
  1. [9]

    Jiangxia Cao, Xixun Lin, Xin Cong, Jing Ya, Tingwen Liu, and Bin Wang. 2022. Disencdr: Learning disentangled representations for cross-domain recommenda- tion. In Proceedings of the 45th International ACM SIGIR conference on research and development in information retrieval . 267–277

    2022

  2. [10]

    Shu Chen, Zitao Xu, Weike Pan, Qiang Yang, and Zhong Ming. 2024. A Survey on Cross-Domain Sequential Recommendation. arXiv preprint arXiv:2401.04971 (2024)

    2024 arXiv

  3. [11]

    Yujia Ding, Huan Li, Ke Chen, and Lidan Shou. 2023. TPUF: Enhancing Cross- domain Sequential Recommendation via Transferring Pre-trained User Features. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management. 410–419

    2023

  4. [12]

    Junchen Fu, Fajie Yuan, Yu Song, Zheng Yuan, Mingyue Cheng, Shenghui Cheng, Jiaqi Zhang, Jie Wang, and Yunzhu Pan. 2024. Exploring adapter-based transfer learning for recommender systems: Empirical studies and practical insights. In Proceedings of the 17th ACM International Co...

    2024

  5. [13]

    Guibing Guo, Jie Zhang, and Neil Yorke-Smith. 2015. Trustsvd: Collaborative filtering with both the explicit and implicit influence of user trust and of item ratings. In Proceedings of the AAAI conference on artificial intelligence , Vol. 29

    2015

  6. [14]

    Lei Guo, Li Tang, Tong Chen, Lei Zhu, Quoc Viet Hung Nguyen, and Hongzhi Yin. 2021. DA-GCN: A domain-aware attentive graph convolution network for shared-account cross-domain sequential recommendation. arXiv preprint arXiv:2105.03300 (2021)

    2021 arXiv

  7. [15]

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685 (2021)

    2021 arXiv

  8. [16]

    Guangneng Hu, Yu Zhang, and Qiang Yang. 2018. Conet: Collaborative cross networks for cross-domain recommendation. In Proceedings of the 27th ACM international conference on information and knowledge management . 667–676

    2018

  9. [17]

    Jun Hu, Wenwen Xia, Xiaolu Zhang, Chilin Fu, Weichang Wu, Zhaoxin Huan, Ang Li, Zuoli Tang, and Jun Zhou. 2024. Enhancing sequential recommendation via llm-based semantic embedding learning. In Companion Proceedings of the ACM on Web Conference 2024. 103–111

    2024

  10. [18]

    Zhenhua Huang, Xin Xu, Juan Ni, Honghao Zhu, and Cheng Wang. 2019. Multi- modal representation learning for recommendation in Internet of Things. IEEE Internet of Things Journal 6, 6 (2019), 10675–10685

    2019

  11. [19]

    Mohsen Jamali and Martin Ester. 2010. A matrix factorization technique with trust propagation for recommendation in social networks. In Proceedings of the fourth ACM conference on Recommender systems . 135–142

    2010

  12. [20]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recom- mendation. In 2018 IEEE International Conference on Data Mining (ICDM) . IEEE, 197–206

    2018

  13. [21]

    Yehuda Koren. 2008. Factorization meets the neighborhood: a multifaceted collaborative filtering model. InProceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining . 426–434

    2008

  14. [22]

    Chenglin Li, Mingjun Zhao, Huanming Zhang, Chenyun Yu, Lei Cheng, Guoqiang Shu, Beibei Kong, and Di Niu. 2022. RecGURU: Adversarial learning of generalized user representations for cross-domain recommendation. In Proceedings of the fifteenth ACM international conference on web...

    2022

  15. [23]

    Pan Li, Zhichao Jiang, Maofei Que, Yao Hu, and Alexander Tuzhilin. 2021. Dual attentive sequential learning for cross-domain click-through rate prediction. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. 3172–3180

    2021

  16. [24]

    Pan Li and Alexander Tuzhilin. 2020. Ddtcdr: Deep dual transfer cross domain recommendation. In Proceedings of the 13th International Conference on Web Search and Data Mining . 331–339

    2020

  17. [25]

    Xiaopeng Li, Fan Yan, Xiangyu Zhao, Yichao Wang, Bo Chen, Huifeng Guo, and Ruiming Tang. 2023. Hamur: Hyper adapter for multi-domain recommendation. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management. 1268–1277

    2023

  18. [26]

    Jiayi Liao, Sihang Li, Zhengyi Yang, Jiancan Wu, Yancheng Yuan, Xiang Wang, and Xiangnan He. 2024. Llara: Large language-recommendation assistant. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval . 1785–1795

    2024

  19. [27]

    Guanyu Lin, Chen Gao, Yu Zheng, Jianxin Chang, Yanan Niu, Yang Song, Kun Gai, Zhiheng Li, Depeng Jin, Yong Li, et al . 2024. Mixed attention network for cross-domain sequential recommendation. In Proceedings of the 17th ACM International Conference on Web Search and Data Minin...

    2024

  20. [28]

    Xinyu Lin, Wenjie Wang, Yongqi Li, Shuo Yang, Fuli Feng, Yinwei Wei, and Tat- Seng Chua. 2024. Data-efficient Fine-tuning for LLM-based Recommendation. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval . 365–374

    2024

  21. [29]

    Yang Lin, Xinyu Ma, Xu Chu, Yujie Jin, Zhibang Yang, Yasha Wang, and Hong Mei. 2024. Lora dropout as a sparsity regularizer for overfitting control. arXiv preprint arXiv:2404.09610 (2024)

    2024 arXiv

  22. [30]

    Meng Liu, Jianjun Li, Guohui Li, and Peng Pan. 2020. Cross domain recom- mendation via bi-directional transfer graph collaborative filtering networks. In Proceedings of the 29th ACM international conference on information & knowledge management. 885–894

    2020

  23. [31]

    Weiming Liu, Xiaolin Zheng, Chaochao Chen, Jiajie Su, Xinting Liao, Mengling Hu, and Yanchao Tan. 2023. Joint internal multi-interest exploration and external domain alignment for cross domain sequential recommendation. In Proceedings of the ACM Web Conference 2023 . 383–394

    2023

  24. [32]

    Haokai Ma, Ruobing Xie, Lei Meng, Xin Chen, Xu Zhang, Leyu Lin, and Jie Zhou. 2024. Triple sequence learning for cross-domain recommendation. ACM Transactions on Information Systems 42, 4 (2024), 1–29

    2024

  25. [33]

    Hao Ma, Haixuan Yang, Michael R Lyu, and Irwin King. 2008. Sorec: social recommendation using probabilistic matrix factorization. In Proceedings of the 17th ACM conference on Information and knowledge management . 931–940

    2008

  26. [34]

    Muyang Ma, Pengjie Ren, Zhumin Chen, Zhaochun Ren, Lifan Zhao, Peiyu Liu, Jun Ma, and Maarten de Rijke. 2022. Mixed information flow for cross-domain sequential recommendations. ACM Transactions on Knowledge Discovery from Data (TKDD) 16, 4 (2022), 1–32

    2022

  27. [35]

    2019.𝜋-net: A parallel information-sharing network for shared-account cross- domain sequential recommendations

    Muyang Ma, Pengjie Ren, Yujie Lin, Zhumin Chen, Jun Ma, and Maarten de Rijke. 2019.𝜋-net: A parallel information-sharing network for shared-account cross- domain sequential recommendations. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Developme...

    2019

  28. [36]

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

  29. [37]

    Chung Park, Taesan Kim, Taekyoon Choi, Junui Hong, Yelim Yu, Mincheol Cho, Kyunam Lee, Sungil Ryu, Hyungjun Yoon, Minsung Choi, et al. 2023. Cracking the Code of Negative Transfer: A Cooperative Game Theoretic Approach for Cross-Domain Sequential Recommendation. In Proceedings...

    2023

  30. [38]

    Gustavo Penha and Claudia Hauff. 2020. What does bert know about books, movies and music? probing bert for conversational recommendation. In Proceed- ings of the 14th ACM conference on recommender systems . 388–397

    2020

  31. [39]

    Aleksandr Petrov and Craig Macdonald. 2022. gSASRec: Reducing Overcon- fidence in Sequential Recommendation Trained with Negative Sampling. In Proceedings of the 17th ACM Conference on Recommender Systems

    2022

  32. [40]

    Tianshu Shen, Jiaru Li, Mohamed Reda Bouadjenek, Zheda Mai, and Scott Sanner

  33. [41]

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

  34. [42]

    Wenchao Sun, Muyang Ma, Pengjie Ren, Yujie Lin, Zhumin Chen, Zhaochun Ren, Jun Ma, and Maarten De Rijke. 2021. Parallel split-join networks for shared account cross-domain sequential recommendations. IEEE Transactions on Knowl- edge and Data Engineering (2021)

    2021

  35. [43]

    Jiaxi Tang and Ke Wang. 2018. Personalized top-n sequential recommendation via convolutional sequence embedding. In Proceedings of the 11th ACM International Conference acronym ’XX, June 03–05, 2018, Woodstock, NY Qingtian Bian, Marcus Vinícius de Carvalho, Tieying Li, Jiaxing...

    2018

  36. [44]

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yas- mine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhos- ale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288 (2023)

    2023 arXiv

  37. [45]

    Aäron van den Oord, Yazhe Li, and Oriol Vinyals. 2018. Representation Learning with Contrastive Predictive Coding. CoRR abs/1807.03748 (2018). arXiv:1807.03748 http://arxiv.org/abs/1807.03748

    2018 arXiv

  38. [46]

    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 systems 30 (2017)

    2017

  39. [47]

    Tianxin Wang, Fuzhen Zhuang, Zhiqiang Zhang, Daixin Wang, Jun Zhou, and Qing He. 2021. Low-dimensional alignment for cross-domain recommendation. In Proceedings of the 30th ACM international conference on information & knowledge management. 3508–3512

    2021

  40. [48]

    Wei Wei, Chao Huang, Lianghao Xia, and Chuxu Zhang. 2023. Multi-modal self-supervised learning for recommendation. In Proceedings of the ACM Web Conference 2023. 790–800

    2023

  41. [49]

    Yinwei Wei, Xiang Wang, Liqiang Nie, Xiangnan He, Richang Hong, and Tat-Seng Chua. 2019. MMGCN: Multi-modal graph convolution network for personalized recommendation of micro-video. In Proceedings of the 27th ACM international conference on multimedia. 1437–1445

    2019

  42. [50]

    Zitao Xu, Weike Pan, and Zhong Ming. 2023. A Multi-view Graph Contrastive Learning Framework for Cross-Domain Sequential Recommendation. In Proceed- ings of the 17th ACM Conference on Recommender Systems . 491–501

    2023

  43. [51]

    Bowen Yang, Cong Han, Yu Li, Lei Zuo, and Zhou Yu. 2021. Improving Con- versational Recommendation Systems’ Quality with Context-Aware Item Meta Information. arXiv preprint arXiv:2112.08140 (2021)

    2021 arXiv

  44. [52]

    Xiaoxin Ye, Yun Li, and Lina Yao. 2023. DREAM: Decoupled Representation via Extraction Attention Module and Supervised Contrastive Learning for Cross- Domain Sequential Recommender. In Proceedings of the 17th ACM Conference on Recommender Systems. 479–490

    2023

  45. [53]

    Bin Yin, Junjie Xie, Yu Qin, Zixiang Ding, Zhichao Feng, Xiang Li, and Wei Lin

  46. [54]

    Zizhuo Zhang and Bang Wang. 2023. Prompt learning for news recommendation. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval . 227–237

    2023

  47. [55]

    Chuang Zhao, Hongke Zhao, Ming He, Jian Zhang, and Jianping Fan. 2023. Cross- domain recommendation via user interest alignment. In Proceedings of the ACM Web Conference 2023. 887–896

    2023

  48. [56]

    Jujia Zhao, Wenjie Wang, Chen Xu, Zhaochun Ren, See-Kiong Ng, and Tat- Seng Chua. 2024. Llm-based federated recommendation. arXiv preprint arXiv:2402.09959 (2024)

    2024 arXiv

  49. [57]

    In Proceedings of the 17th ACM Conference on Recommender Systems

    Heterogeneous knowledge fusion: A novel approach for personalized rec- ommendation via llm. In Proceedings of the 17th ACM Conference on Recommender Systems. 599–601

  50. [58]

    Feng Zhu, Chaochao Chen, Yan Wang, Guanfeng Liu, and Xiaolin Zheng. 2019. Dtcdr: A framework for dual-target cross-domain recommendation. In Proceed- ings of the 28th ACM international conference on information and knowledge management. 1533–1542

    2019

  51. [59]

    Feng Zhu, Yan Wang, Chaochao Chen, Guanfeng Liu, and Xiaolin Zheng. 2020. A graphical and attentional framework for dual-target cross-domain recommen- dation.. In IJCAI, Vol. 21. 39

    2020

  52. [60]

    Yongchun Zhu, Zhenwei Tang, Yudan Liu, Fuzhen Zhuang, Ruobing Xie, Xu Zhang, Leyu Lin, and Qing He. 2022. Personalized transfer of user preferences for cross-domain recommendation. In Proceedings of the fifteenth ACM international conference on web search and data mining . 1507–1515

    2022

  53. [61]

    Xiaolin Zheng, Jiajie Su, Weiming Liu, and Chaochao Chen. 2022. DDGHM: dual dynamic graph with hybrid metric training for cross-domain sequential recommendation. In Proceedings of the 30th ACM International Conference on Multimedia. 471–481

    2022

  54. [2015]

    In Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval

    Image-based recommendations on styles and substitutes. In Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval. 43–52

  55. [2019]

    In Proceedings of the 28th ACM International conference on Information and Knowledge Management

    BERT4Rec: Sequential recommendation with bidirectional encoder rep- resentations from transformer. In Proceedings of the 28th ACM International conference on Information and Knowledge Management . 1441–1450

  56. [2023]

    Information Processing & Man- agement 60, 1 (2023), 103139

    Towards understanding and mitigating unintended biases in language model-driven conversational recommendation. Information Processing & Man- agement 60, 1 (2023), 103139

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.