Pith. sign in

REVIEW 3 major objections 5 minor 54 references

Cross-domain recommenders improve when they model how user tastes shift across categories inside a domain and transfer only the matching pieces.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-15 13:28 UTC pith:6PMHIWT7

load-bearing objection Solid empirical CDR recipe with real IDH evidence and public code; the causal story for multi-criteria + doppelganger is only partly isolated, but the work still deserves referees. the 3 major comments →

arxiv 2603.07086 v2 pith:6PMHIWT7 submitted 2026-03-07 cs.HC cs.IR

Multi-TAP: Multi-criteria Target Adaptive Persona Modeling for Cross-Domain Recommendation

Daehee Kang , Yeon-Chang Lee This is my paper
classification cs.HC cs.IR
keywords cross-domain recommendationintra-domain heterogeneitypersona modelinglarge language modelknowledge transferrecommender systems
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Cross-domain recommendation borrows a user’s history from one product area to help recommend in another when data is sparse. Most methods treat each domain as if the user has one stable preference summary, then map or align that summary across domains. This paper shows that even inside one domain—electronics, home, sports—users change how they shop across categories: they may buy expensive computers but cheap audio gear, and they weight different criteria such as price, popularity, quality, category familiarity, and diversity. Multi-TAP therefore builds several criterion-specific natural-language personas per user, encodes them into embeddings, and transfers only a target-conditioned “doppelganger” of those personas rather than the whole source profile. On real Amazon domain pairs this consistently beats strong single-domain and cross-domain baselines on ranking metrics, arguing that modeling that within-domain heterogeneity is necessary for reliable transfer.

Core claim

Existing cross-domain methods compress user behavior into domain-level representations and thereby erase context-dependent preference signals that already vary across categories inside a single domain. Explicit multi-criteria persona modeling plus selective, target-adaptive doppelganger transfer preserves those signals and yields higher top-k accuracy than state-of-the-art CDR methods on Amazon domain pairs, with gains up to roughly one-third on hit rate in the hardest reported setting.

What carries the argument

Multi-TAP: five criterion-specific personas (price sensitivity, quality preference, popularity bias, category diversity, category familiarity) built from category-wise ordinal labels, verbalized by an LLM under a fixed structured prompt, encoded and self-attention-aggregated into a domain persona embedding; plus a target-initialized doppelganger that absorbs source signals via cross-attention and is aligned to the target persona by InfoNCE only on the target side.

Load-bearing premise

The method assumes that five fixed preference criteria, discretized into low/medium/high labels per category and written as personas by a language model under a fixed prompt, faithfully capture the preference pieces that should be transferred across domains.

What would settle it

Measure category-level preference-preservation ratios on held-out domain pairs; if Multi-TAP loses its margin over direct source–target alignment precisely where preservation is high (low intra-domain heterogeneity), or if a single-persona model using the same processed metadata matches Multi-TAP’s HR@5 and NDCG@5 on the paper’s Amazon splits, the claim that multi-criteria modeling of intra-domain heterogeneity drives the gains would fail.

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

If this is right

  • Domain-level user embeddings that ignore category-level preference shifts lose information before any cross-domain map is learned.
  • Selective persona-level transfer can outperform both pure single-domain graph models and direct source–target embedding alignment.
  • Users with very few target interactions can still benefit when source signals are filtered by target relevance.
  • Relative ordinal preference labels per category are more useful for transfer than raw absolute metadata values.
  • Persona generation can stay offline while the online model remains a light add-on to standard collaborative-filtering backbones.

Where Pith is reading between the lines

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

  • The same criterion decomposition could help multi-behavior or multi-session recommenders where context changes inside one platform.
  • When price or rating metadata are missing, alternative behavioral proxies for the five criteria would need separate validation before the method transfers.
  • Doppelganger-style intermediate anchors may reduce negative transfer in other multi-domain representation tasks beyond retail.
  • Relative gains should shrink when source and target share little category structure—a natural stress test the paper leaves open.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper argues that cross-domain recommendation (CDR) is limited by the assumption of homogeneous preferences even within a single domain, and documents intra-domain heterogeneity (IDH) via category-level conditional preservation ratios on Amazon data (price, popularity, etc.). It proposes Multi-TAP: (P1) multi-criteria persona modeling that builds a structured user persona DB from five criteria (price sensitivity, quality, popularity, category diversity/familiarity), verbalizes them with a constrained LLM prompt into five natural-language personas, encodes them, and aggregates via self-attention; (P2) target-adaptive doppelganger transfer that initializes a target-centered embedding, refines it by cross-attention to the source persona, and aligns via InfoNCE, jointly with BPR on LightGCN+semantic item embeddings. On time-aware Amazon domain pairs under full ranking, Multi-TAP reports consistent gains over single-domain and CDR baselines (up to 36.3% HR@5 / 42.6% NDCG@5 on Home o Elec), with ablations on multi- vs single-persona, aggregation, cold-start, direct vs doppelganger transfer, hyperparameters, and LLM variants, plus a public codebase.

Significance. If the causal story holds, the work is a useful contribution to CDR: it elevates IDH from a side observation to a design principle, couples structured multi-criteria signals with LLM verbalization under a controlled prompt, and offers a target-centered transfer alternative to direct source–target collapse. Strengths include multi-faceted empirical support (IDH heatmaps in Fig. 2/6; full-ranking HR/NDCG with means±std over five seeds in Tables 4/12; ablations in Tables 5–7, 10–11 and Figs. 4–5), a time-aware split that is more realistic than random sampling, and released code. Absolute HR@5 remains low (~1–4%), so the practical impact is incremental rather than transformative, but the framing and selective-transfer design are of clear interest to the recommender-systems community.

major comments (3)
  1. §4.2 EQ1–EQ3 and Tables 5, 11, Fig. 4: The central claim attributes lifts to multi-criteria IDH modeling plus doppelganger transfer, but the ablations do not cleanly isolate that attribution. Multi vs single persona (Table 5) yields only modest deltas (e.g., Sports o Cloth HR 3.77 o4.09); removing LLM personas (Table 11) causes a large drop (Toys o Home HR 3.11 o1.87), leaving open whether gains are mainly semantic enrichment of LightGCN rather than the five-criteria IDH story; and there is no full factorial that keeps multi-criteria personas while removing only the doppelganger path (or vice versa) across the six main pairs. A complete ablation matrix (or at least multi-persona + direct transfer vs multi-persona + doppelganger on all main pairs) is needed to support the headline causal claim.
  2. §2 Measurement and §3.2 Steps 1–2 (Eqs. 1–8, Table 2): The five hand-chosen criteria, category-wise tertile discretization, and fixed structured LLM verbalization are treated as sufficient carriers of transferable preference. The paper does not show that these criteria are necessary or complete (e.g., sensitivity to alternative criteria sets, continuous vs tertile signals, or non-LLM encoding of the same ordinal labels). Without that, the claim that Multi-TAP works by “explicitly modeling IDH” rests on an untested sufficiency assumption; at minimum, a criteria-ablation or non-LLM structured-feature baseline beyond the partial Table 11 condition should be reported for the main pairs.
  3. Table 4 Cloth o Sports and several reverse directions in Table 12: Multi-TAP is not uniformly best (second to PPA on Cloth o Sports; mixed on some other pairs). The abstract and EQ1 language (“consistently outperforms”) should be qualified, and the paper should analyze when doppelganger transfer helps vs hurts (e.g., domain size, overlap, or IDH strength) rather than averaging over wins.
minor comments (5)
  1. Notation in §3 is dense and occasionally inconsistent (e.g., doppelganger symbols and domain subscripts in Eqs. 12–14); a short notation checklist or simplified diagram callouts would help.
  2. Absolute metrics are low under full ranking; briefly discuss practical significance and whether top-k or sampled ranking would change relative conclusions.
  3. §4.1: Cost (~$0.022/user, ~3s/user offline) is useful; state whether personas are frozen after one pass or refreshed, and how cold users without enough history are handled.
  4. Figure 1 vs Figure 2: clarify the relationship between “preference shift ratios” and “conditional preservation ratios” so readers can map the motivation figure to the formal measure.
  5. Related work §5: briefly contrast with multi-interest / multi-prototype CDR beyond PPA (e.g., REMIT, DisCo) on whether criteria are explicit vs latent.

Circularity Check

0 steps flagged

No circular derivation chain; Multi-TAP is an empirical CDR architecture whose ranking lifts are measured on held-out interactions, not quantities forced by construction from its own inputs.

full rationale

The paper's central claims are architectural (multi-criteria LLM-assisted persona construction from category-wise ordinal labels + target-centered doppelganger contrastive transfer) and empirical (HR/NDCG gains vs. baselines under a time-aware Amazon split). Section 2 defines and measures IDH via preservation ratios on observed interactions; those ratios motivate the five hand-chosen criteria but are never algebraically equated to the later recommendation scores or claimed as 'predicted' by the model. Persona embeddings (Eqs. 9–11) are generated from structured metadata and encoded once; doppelganger refinement (Eq. 12) and InfoNCE (Eq. 13) are ordinary contrastive objectives jointly optimized with BPR (Eqs. 15–16). Hyper-parameters λ and τ are tuned on validation data in the standard way; final metrics are computed on held-out post-2019 interactions. No equation reduces a reported lift to a fitted constant, no uniqueness theorem is imported from the authors' prior work, and no self-citation is load-bearing for the performance claim. Ablation incompleteness (Tables 5–7, Fig. 4) is an attribution concern, not circularity. The derivation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 2 invented entities

The work is an empirical systems paper. Its load-bearing commitments are modeling choices (five criteria, tertile discretization, LLM verbalization, doppelganger construction) and standard recsys losses rather than free physical constants or unproved mathematical lemmas. The free parameters are ordinary hyper-parameters; the invented entities are the multi-persona and doppelganger constructs introduced to operationalize IDH.

free parameters (5)
  • lambda (weight of doppelganger InfoNCE)
    Controls contribution of transfer loss; tuned on validation; non-trivial values improve HR@5 (Fig. 5).
  • tau (InfoNCE temperature)
    Temperature for contrastive alignment; stable in a moderate range but still a free choice (Fig. 5).
  • number and identity of preference criteria K
    Fixed to five hand-selected criteria (PS, QP, PB, CD, CF); not learned from data.
  • tertile binning thresholds
    Category-wise 1/3 and 2/3 quantiles used to create Low/Medium/High labels (Eq. 1, 5, 7).
  • embedding dimensions and LightGCN depth
    User/item dim 128, persona encoding dim, item semantic 512; standard but fitted choices.
axioms (4)
  • domain assumption User preferences within a domain are heterogeneous across categories and can be usefully summarized by a small set of ordinal criteria derived from price, rating, popularity, and category statistics.
    Stated as the core motivation and formalized in §2; supported by heatmaps but still an assumption about what matters for transfer.
  • domain assumption An LLM under a constrained structured prompt can faithfully verbalize the ordinal criterion labels into persona text without introducing harmful hallucination.
    §3.2 Step 2 and Table 2; ablation without LLM (Table 11) shows drop but does not prove faithfulness of the generated text.
  • ad hoc to paper Target-centered doppelganger alignment (copy + cross-attention + InfoNCE) transfers useful source signal without collapsing domain-specific preferences.
    Design choice of (P2); justified by comparison to direct alignment (Fig. 4) rather than theory.
  • domain assumption BPR on LightGCN ID embeddings plus semantic item text is an adequate collaborative backbone for the final ranking.
    Standard recsys practice adopted in §3.4.
invented entities (2)
  • Multi-criteria user persona embeddings (five criterion-specific LLM texts aggregated by self-attention) no independent evidence
    purpose: Explicitly represent intra-domain heterogeneity instead of a single domain-level user vector.
    Core construct of (P1); independent evidence is only the downstream ranking gains and ablations, not an external measurement of persona fidelity.
  • Target-adaptive doppelganger persona embedding no independent evidence
    purpose: Intermediate target-centered vessel that selectively absorbs source signals before contrastive alignment.
    Introduced in §3.3; no existence claim outside the model; evidence is comparative performance vs direct InfoNCE.

pith-pipeline@v1.1.0-grok45 · 25016 in / 3198 out tokens · 29580 ms · 2026-07-15T13:28:11.826515+00:00 · methodology

0 comments
read the original abstract

Cross-domain recommendation (CDR) aims to alleviate data sparsity by transferring knowledge across domains, yet existing methods primarily rely on coarse-grained behavioral signals and often overlook intra-domain heterogeneity in user preferences. We propose Multi-TAP, a multi-criteria target-adaptive persona framework that explicitly captures such heterogeneity through semantic persona modeling. To enable effective transfer, Multi-TAP selectively incorporates source-domain signals conditioned on the target domain, preserving relevance during knowledge transfer. Experiments on real-world datasets demonstrate that Multi-TAP consistently outperforms state-of-the-art CDR methods, highlighting the importance of modeling intra-domain heterogeneity for robust cross-domain recommendation. The codebase of Multi-TAP is currently available at https://github.com/archivehee/Multi-TAP.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references · 7 linked inside Pith

  1. [1]

    Icek Ajzen. 1991. The theory of planned behavior.Organizational Behavior and Human Decision Processes50, 2 (1991)

    1991

  2. [3]

    InProceedings of the 26th International Joint Conference on Artificial Intelligence

    Cross-Domain Recommendation: An Embedding and Mapping Approach. InProceedings of the 26th International Joint Conference on Artificial Intelligence

  3. [4]

    Jiangxia Cao, Shaoshuai Li, Bowen Yu, Xiaobo Guo, Tingwen Liu, and Bin Wang

  4. [5]

    InProceedings of the 16th ACM international conference on web search and data mining

    Towards Universal Cross-Domain Recommendation. InProceedings of the 16th ACM international conference on web search and data mining

  5. [6]

    Jiangxia Cao, Xixun Lin, Xin Cong, Jing Ya, Tingwen Liu, and Bin Wang. 2022. DisenCDR: Learning Disentangled Representations for Cross-Domain Recom- mendation. InProceedings of the 15th ACM conference on Recommender systems

    2022

  6. [7]

    Jiangxia Cao, Jiawei Sheng, Xin Cong, Tingwen Liu, and Bin Wang. 2022. Cross- domain recommendation to cold-start users via variational information bottle- neck. InProceedings of the IEEE 38th International Conference on data engineering (ICDE)

    2022

  7. [8]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A Simple Framework for Contrastive Learning of Visual Representations. In Proceedings of the 37th International Conference on Machine Learning

    2020

  8. [9]

    Yue Chen, Susen Yang, Tong Zhang, Chao Wang, Mingyue Cheng, Chenyi Lei, and Han Li. 2025. Lasso: Large Language Model-based User Simulator for Cross- Domain Recommendation. InProceedings of the 19th ACM Conference on Recom- mender Systems

    2025

  9. [10]

    Manuel Cossio. 2025. A comprehensive taxonomy of hallucinations in large language models.arXiv preprint arXiv:2508.01781(2025)

    Pith/arXiv arXiv 2025

  10. [11]

    Jing Du, Zesheng Ye, Bin Guo, Zhiwen Yu, and Lina Yao. 2023. Distributional Domain-Invariant Preference Matching for Cross-Domain Recommendation. In Proceedings of the 23rd IEEE International Conference on Data Mining (ICDM)

    2023

  11. [12]

    Bowen Hao, Chaoqun Yang, Lei Guo, Junliang Yu, and Hongzhi Yin. 2024. Motif- based prompt learning for universal cross-domain recommendation. InProceed- ings of the 17th ACM international conference on web search and data mining

    2024

  12. [13]

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

    2020

  13. [14]

    Yupeng Hou, Jiacheng Li, Zhankui He, An Yan, Xiusi Chen, and Julian McAuley

  14. [15]

    Bridging Language and Items for Retrieval and Recommendation.arXiv preprint arXiv:2403.03952(2024)

    Pith/arXiv arXiv 2024

  15. [16]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, et al. 2025. A survey on hallucination in large language models: Principles, taxonomy, chal- lenges, and open questions.ACM Transactions on Information Systems43, 2 (2025)

    2025

  16. [17]

    Hyunjun Ju, SeongKu Kang, Dongha Lee, Junyoung Hwang, Sanghwan Jang, and Hwanjo Yu. 2024. Multi-domain recommendation to attract users via domain preference modeling. InProceedings of the 38th AAAI Conference on Artificial Intelligence

    2024

  17. [18]

    SeongKu Kang, Junyoung Hwang, Dongha Lee, and Hwanjo Yu. 2019. Semi- Supervised Learning for Cross-Domain Recommendation to Cold-Start Users. InProceedings of the 28th ACM International Conference on Information and Knowledge Management

    2019

  18. [19]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-Attentive Sequential Rec- ommendation. InProceedings of the 18th IEEE International Conference on Data Mining (ICDM)

    2018

  19. [20]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. 2014. Adam: A Method for Stochastic Opti- mization.arxiv preprint arXiv.1412.6980(2014)

    Pith/arXiv arXiv 2014

  20. [21]

    Yeon-Chang Lee, JaeHyun Lee, Michiharu Yamashita, Dongwon Lee, and Sang- Wook Kim. 2025. Caper: Enhancing career trajectory prediction using temporal knowledge graph and ternary relationship. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    2025

  21. [22]

    Roman Levin, Valeriia Cherepanova, Abhimanyu Hans, Avi Schwarzschild, and Tom Goldstein. 2025. Has My System Prompt Been Used? Large Language Model Prompt Membership Inference.arXiv preprint arXiv:2502.09974(2025)

    Pith/arXiv arXiv 2025

  22. [23]

    Mosh Levy, Alon Jacoby, and Yoav Goldberg. 2024. Same task, more tokens: the impact of input length on the reasoning performance of large language models. arXiv preprint arXiv:2402.14848(2024)

    Pith/arXiv arXiv 2024

  23. [24]

    Bin Li, Qiang Yang, and Xiangyang Xue. 2009. Can movies and books collaborate? cross-domain collaborative filtering for sparsity reduction.. InProceedings of the 21th International Joint Conference on Artificial Intelligence

    2009

  24. [25]

    Chenglin Li, Yuanzhen Xie, Chenyun Yu, Bo Hu, Zang Li, Guoqiang Shu, Xiaohu Qie, and Di Niu. 2023. One for All, All for One: Learning and Transferring User Embeddings for Cross-Domain Recommendation. InProceedings of the 16th ACM international conference on web search and data mining

    2023

  25. [26]

    Hanyu Li, Jiayu Li, Weizhi Ma, Peijie Sun, Haiyang Wu, Jingwen Wang, Yuekui Yang, Min Zhang, and Shaoping Ma. 2025. CD-CDR: Conditional Diffusion-based Item Generation for Cross-Domain Recommendation. InProceedings of the 48th International ACM SIGIR conference on research and development in Information Retrieval

    2025

  26. [27]

    Hanyu Li, Weizhi Ma, Peijie Sun, Jiayu Li, Cunxiang Yin, Yancheng He, Guoqiang Xu, Min Zhang, and Shaoping Ma. 2024. Aiming at the Target: Filter Collabora- tive Information for Cross-Domain Recommendation. InProceedings of the 47th International ACM SIGIR conference on research and development in Information Retrieval

    2024

  27. [28]

    Hourun Li, Yifan Wang, Zhiping Xiao, Jia Yang, Changling Zhou, Ming Zhang, and Wei Ju. 2025. DisCo: graph-based disentangled contrastive learning for cold- start cross-domain recommendation. InProceedings of the 39th AAAI Conference on Artificial Intelligence

    2025

  28. [29]

    Meng Liu, Jianjun Li, Guohui Li, and Peng Pan. 2020. Cross Domain Recommen- dation via Bi-directional Transfer Graph Collaborative Filtering Networks. In Proceedings of the 29th ACM International Conference on Information and Knowl- edge Management

    2020

  29. [30]

    Wentao Ning, Xiao Yan, Weiwen Liu, Reynold Cheng, Rui Zhang, and Bo Tang

  30. [31]

    InProceedings of the 32nd ACM International Conference on Information and Knowledge Management

    Multi-domain Recommendation with Embedding Disentangling and Do- main Alignment. InProceedings of the 32nd ACM International Conference on Information and Knowledge Management

  31. [32]

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

  32. [33]

    InProceedings of the 25th Conference on Uncertainty in Artificial Intelligence

    BPR: Bayesian personalized ranking from implicit feedback. InProceedings of the 25th Conference on Uncertainty in Artificial Intelligence

  33. [34]

    Aghiles Salah, Thanh-Binh Tran, and Hady W. Lauw. 2021. Towards Source- Aligned Variational Models for Cross-Domain Recommendation. InProceedings of the 15th ACM conference on Recommender systems

    2021

  34. [35]

    Abdulaziz Samra, Evgeny Frolov, Alexey Vasilev, Alexander Grigorevskiy, and Anton Vakhrushev. 2024. Cross-domain latent factors sharing via implicit matrix factorization. InProceedings of the 18th ACM conference on Recommender systems

    2024

  35. [36]

    Singh and Geoffrey J

    Ajit P. Singh and Geoffrey J. Gordon. 2008. Relational learning via collective matrix factorization. InProceedings of the 14th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    2008

  36. [37]

    Zijian Song, Wenhan Zhang, Lifang Deng, Jiandong Zhang, Zhihua Wu, Kaigui Bian, and Bin Cui. 2024. Mitigating negative transfer in cross-domain recom- mendation via knowledge transferability enhancement. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    2024

  37. [38]

    Caiqi Sun, Jiewei Gu, BinBin Hu, Xin Dong, Hai Li, Lei Cheng, and Linjian Mo

  38. [39]

    InProceedings of the 37th AAAI conference on artificial intelligence

    REMIT: reinforced multi-interest transfer for cross-domain recommenda- tion. InProceedings of the 37th AAAI conference on artificial intelligence

  39. [40]

    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

    2019

  40. [41]

    Yuhan Wang, Qing Xie, Zhifeng Bao, Mengzi Tang, Lin Li, and Yongjian Liu. 2025. Enhancing transferability and consistency in cross-domain recommendations via supervised disentanglement. InProceedings of the 19th ACM Conference on Recommender Systems

    2025

  41. [42]

    Wenhao Yang, Yingchun Jian, Yibo Wang, Shiyin Lu, Lei Shen, Bing Wang, Hai- hong Tang, and Lijun Zhang. 2024. Not all embeddings are created equal: towards robust cross-domain recommendation via contrastive learning. InProceedings of the ACM Web Conference

    2024

  42. [43]

    Tianzi Zang, Yanmin Zhu, Haobing Liu, Ruohan Zhang, and Jiadi Yu. 2022. A Survey on Cross-domain Recommendation: Taxonomies, Methods, and Future Directions.ACM Transactions on Information Systems41, 2 (2022)

    2022

  43. [44]

    Tingting Zhang, Pengpeng Zhao, Yanchi Liu, Victor S Sheng, Jiajie Xu, Deqing Wang, Guanfeng Liu, Xiaofang Zhou, et al . 2019. Feature-level deeper self- attention network for sequential recommendation.. InProceedings of the 28th International Joint Conference on Artificial Intelligence

    2019

  44. [45]

    Yu Zhang, Bin Cao, and Dit-Yan Yeung. 2012. Multi-Domain Collaborative Filtering.arXiv preprint arXiv:1203.3535(2012)

    Pith/arXiv arXiv 2012

  45. [46]

    Yuxi Zhang, Ji Zhang, Feiyang Xu, Lvying Chen, Bohan Li, Lei Guo, and Hongzhi Yin. 2024. Preference Prototype-Aware Learning for Universal Cross-Domain Recommendation. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management

    2024

  46. [47]

    Chuang Zhao, Hongke Zhao, Ming He, Jian Zhang, and Jianping Fan. 2023. Cross- domain recommendation via user interest alignment. InProceedings of the ACM Web conference

    2023

  47. [48]

    Yi Zhao, Chaozhuo Li, Jiquan Peng, Xiaohan Fang, Feiran Huang, Senzhang Wang, Xing Xie, and Jibing Gong. 2023. Beyond the overlapping users: Cross-domain recommendation via adaptive anchor link learning. InProceedings of the 46th International ACM SIGIR conference on research and development in Information Retrieval

    2023

  48. [49]

    Donglin Zhou, Xinbei Cai, and Weike Pan. 2025. Contrastive Text-enhanced Transformer for Cross-Domain Sequential Recommendation. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    2025

  49. [50]

    Feng Zhu, Chaochao Chen, Yan Wang, Guanfeng Liu, and Xiaolin Zheng. 2019. DTCDR: A Framework for Dual-Target Cross-Domain Recommendation. InPro- ceedings of the 28th ACM International Conference on Information and Knowledge Management

    2019

  50. [51]

    Feng Zhu, Yan Wang, Chaochao Chen, Guanfeng Liu, and Xiaolin Zheng. 2020. A Graphical and Attentional Framework for Dual-Target Cross-Domain Recom- mendation. InProceedings of the 29th International Joint Conference on Artificial Intelligence

    2020

  51. [52]

    Feng Zhu, Yan Wang, Chaochao Chen, Jun Zhou, Longfei Li, and Guanfeng Liu

  52. [53]

    Cross-domain recommendation: challenges, progress, and prospects.arXiv preprint arXiv:2103.01696(2021)

    Pith/arXiv arXiv 2021

  53. [54]

    Yongchun Zhu, Kaikai Ge, Fuzhen Zhuang, Ruobing Xie, Dongbo Xi, Xu Zhang, Leyu Lin, and Qing He. 2021. Transfer-Meta Framework for Cross-domain Recommendation to Cold-Start Users. InProceedings of the 44th International ACM SIGIR conference on research and development in Information Retrieval. ���������� �������������� ������ �������� ������� �������� ���...

    2021

  54. [55]

    User ID":

    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. InProceedings of the 15th ACM international conference on web search and data mining. A Additional Details of Multi-TAP A.1 User Persona Database Table 8presents a sample from ...

    2022