pith. sign in

arxiv: 2601.02366 · v2 · submitted 2025-11-25 · 💻 cs.IR · cs.AI

TextBridgeGNN: Pre-training Graph Neural Network for Cross-Domain Recommendation via Text-Guided Transfer

Yiwen Chen , Yiqing Wu , Huishi Luo , Fuzhen Zhuang , Deqing Wang , Zhao Zhang This is my paper

Pith reviewed 2026-05-17 05:20 UTC · model grok-4.3

classification 💻 cs.IR cs.AI
keywords cross-domain recommendationgraph neural networkspre-trainingtext-guided transfersemantic bridgecollaborative filteringID embeddingsmulti-domain recommendation
0
0 comments X

The pith

Text serves as a semantic bridge that lets pre-trained graph neural networks transfer ID embeddings and collaborative patterns to new recommendation domains.

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

The paper tries to overcome the fact that standard ID-based graph recommendation models cannot move from one domain to another because their embeddings and graph structures are domain-specific. It does this by treating item text as a common language that connects isolated domains during pre-training with hierarchical graph networks, then uses text similarity to initialize embeddings in a target domain during fine-tuning. A sympathetic reader would care because this removes the need to rebuild models from scratch for each new domain or to keep running heavy language models at inference time, potentially making cross-domain and training-free recommendation practical.

Core claim

TextBridgeGNN builds a pre-training and fine-tuning framework in which textual information breaks the isolation between domains. Hierarchical GNNs are trained to learn both domain-specific and domain-global knowledge while retaining collaborative signals. A similarity transfer mechanism then copies ID embeddings from semantically related source nodes to initialize the target domain, thereby moving both the embeddings and the learned graph patterns without requiring costly language-model updates.

What carries the argument

The similarity transfer mechanism that matches nodes across domains by their text features and initializes target ID embeddings from the most similar source nodes.

If this is right

  • Cross-domain recommendation becomes feasible without rebuilding graph structures or retraining from random ID embeddings in the new domain.
  • Multi-domain pre-training can pool collaborative signals from several sources while still preserving each domain's unique patterns through hierarchical propagation.
  • Training-free deployment in new domains is possible by direct embedding transfer guided by text similarity.
  • Pre-trained language model semantics can be injected into graph collaborative filtering without ongoing fine-tuning or inference-time language model calls.

Where Pith is reading between the lines

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

  • The same text-bridge idea might reduce cold-start problems inside a single domain when new items arrive with descriptive text.
  • If text similarity proves robust, the framework could be tested on non-recommendation graph tasks such as knowledge-graph completion across different ontologies.
  • Future work could examine whether other side information such as images or tags can substitute for text when descriptions are sparse.

Load-bearing premise

Text descriptions of items are consistent enough across domains to serve as a reliable indicator of which nodes should share transferred embeddings and graph patterns.

What would settle it

Measuring whether recommendation accuracy in a target domain drops to the level of random embedding initialization when the source and target domains share little overlapping item text.

Figures

Figures reproduced from arXiv: 2601.02366 by Deqing Wang, Fuzhen Zhuang, Huishi Luo, Yiqing Wu, Yiwen Chen, Zhao Zhang.

Figure 1
Figure 1. Figure 1: Model Architecture: TextBridgeGNN consists of two main phases: (1) Multi-domain pre-training, which employs [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Training-free results on Automotive, Tools, Cell Phones, Clothing, Electronics, Home, Movies → Sports dataset [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Embedding T-SNE in Books & Clothes Domain [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
read the original abstract

Graph-based recommendation has achieved great success in recent years. The classical graph recommendation model utilizes ID embedding to store essential collaborative information. However, this ID-based paradigm faces challenges in transferring to a new domain, making it hard to build a pre-trained graph recommendation model. This phenomenon primarily stems from two inherent challenges: (1) the non-transferability of ID embeddings due to isolated domain-specific ID spaces, and (2) structural incompatibility between heterogeneous interaction graphs across domains. To address these issues, we propose TextBridgeGNN, a pre-training and fine-tuning framework that can effectively transfer knowledge from a pre-trained GNN to downstream tasks. We believe the key lies in how to build the relationship between domains. Specifically, TextBridgeGNN uses text as a semantic bridge to connect domains through multi-level graph propagation. During the pre-training stage, textual information is utilized to break the data islands formed by multiple domains, and hierarchical GNNs are designed to learn both domain-specific and domain-global knowledge with text features, ensuring the retention of collaborative signals and the enhancement of semantics. During the fine-tuning stage, a similarity transfer mechanism is proposed. This mechanism initializes ID embeddings in the target domain by transferring from semantically related nodes, successfully transferring the ID embeddings and graph pattern. Experiments demonstrate that TextBridgeGNN outperforms existing methods in cross-domain, multi-domain, and training-free settings, highlighting its ability to integrate Pre-trained Language Model (PLM)-driven semantics with graph-based collaborative filtering without costly language model fine-tuning or real-time inference overhead.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes TextBridgeGNN, a pre-training and fine-tuning framework for graph-based recommendation that uses text as a semantic bridge to address non-transferability of ID embeddings and structural incompatibility across domains. Hierarchical GNNs learn domain-specific and domain-global knowledge with text features during pre-training; a similarity transfer mechanism then initializes target-domain ID embeddings from semantically related source nodes. Experiments claim outperformance over existing methods in cross-domain, multi-domain, and training-free settings by integrating PLM semantics with collaborative filtering without costly fine-tuning or inference overhead.

Significance. If validated, the approach could meaningfully advance cross-domain recommendation by enabling pre-trained GNNs to transfer both embeddings and graph patterns via text without PLM fine-tuning or real-time overhead. The integration of hierarchical propagation with similarity-based initialization offers a practical route to break domain isolation while retaining collaborative signals.

major comments (2)
  1. [Abstract / fine-tuning stage] Abstract / fine-tuning stage description: the similarity transfer mechanism initializes target-domain ID embeddings by copying from source nodes selected via text-feature similarity. This step is load-bearing for the central claim of successful transfer of ID embeddings and graph patterns, yet the description provides no corrective term or alignment loss for cases where textual overlap is high but interaction densities or user behaviors differ across domains.
  2. [Abstract] Abstract: the claim that TextBridgeGNN 'successfully transferring the ID embeddings and graph pattern' rests on an untested correlation between PLM text embeddings and latent collaborative structure learned by the hierarchical GNNs. No analysis or ablation is referenced that isolates whether performance gains survive when text similarity and graph patterns diverge.
minor comments (2)
  1. [Abstract] Abstract: 'multi-level graph propagation' is mentioned but not defined; clarify how the hierarchical GNNs implement domain-specific versus domain-global layers and where text features are injected.
  2. [Abstract] Abstract: the training-free setting is highlighted as a strength but lacks a brief statement of what 'training-free' concretely means for the target domain (e.g., zero-shot embedding initialization only).

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and detailed comments. We address each major point below, clarifying the current manuscript content and indicating where revisions will be made to improve clarity and rigor.

read point-by-point responses

  1. Referee: [Abstract / fine-tuning stage] Abstract / fine-tuning stage description: the similarity transfer mechanism initializes target-domain ID embeddings by copying from source nodes selected via text-feature similarity. This step is load-bearing for the central claim of successful transfer of ID embeddings and graph patterns, yet the description provides no corrective term or alignment loss for cases where textual overlap is high but interaction densities or user behaviors differ across domains.

    Authors: We agree that the abstract description is concise and does not explicitly address potential mismatches between textual similarity and collaborative behavior. In the full manuscript (Section 3.3), the similarity transfer selects source nodes via PLM embeddings and copies their ID embeddings, after which standard fine-tuning on the target domain adapts the model to local interaction densities. No explicit corrective alignment loss is present during the transfer step itself. We will revise the abstract and add a dedicated paragraph in the fine-tuning section discussing this limitation and how target-domain fine-tuning mitigates discrepancies. revision: yes

  2. Referee: [Abstract] Abstract: the claim that TextBridgeGNN 'successfully transferring the ID embeddings and graph pattern' rests on an untested correlation between PLM text embeddings and latent collaborative structure learned by the hierarchical GNNs. No analysis or ablation is referenced that isolates whether performance gains survive when text similarity and graph patterns diverge.

    Authors: The cross-domain and multi-domain experiments show consistent gains, providing empirical support for the overall framework. However, we acknowledge that the manuscript does not include a targeted ablation or analysis that isolates performance when text similarity and graph patterns diverge. We will add such an analysis in the revised version, including an ablation study on scenarios with controlled divergence (e.g., via dataset subsets or synthetic perturbations) to directly test the robustness of the transfer. revision: yes

Circularity Check

0 steps flagged

No circularity: framework introduces independent mechanisms without reduction to inputs or self-citations

full rationale

The paper's derivation introduces TextBridgeGNN as a new pre-training/fine-tuning framework that explicitly constructs a text-based semantic bridge, designs hierarchical GNNs to capture domain-specific and domain-global knowledge, and defines a similarity transfer mechanism to initialize target ID embeddings from source nodes. These steps are presented as novel proposals to address non-transferability of ID embeddings and structural incompatibility, with no equations or definitions that reduce the output (e.g., transferred embeddings or learned patterns) back to the inputs by construction. No self-citation load-bearing steps, uniqueness theorems, or ansatzes imported from prior author work appear in the abstract or description; performance is instead tied to external experiments. The central claims therefore remain self-contained and do not collapse into fitted parameters or renamed inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework relies on standard assumptions in GNNs and PLMs but introduces the text-bridge concept as a key enabler; details on any fitted parameters are not specified in the abstract.

axioms (1)
  • domain assumption Textual features from pre-trained language models can effectively align and bridge nodes across different recommendation domains
    Central to breaking data islands and enabling transfer.

pith-pipeline@v0.9.0 · 5600 in / 1080 out tokens · 55952 ms · 2026-05-17T05:20:53.869343+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

57 extracted references · 57 canonical work pages · 1 internal anchor

  1. [1]

    AI@Meta. 2024. Llama 3 Model Card. (2024). https://github.com/meta-llama/ llama3/blob/main/MODEL_CARD.md

    work page 2024

  2. [2]

    Jianxin Chang, Chenbin Zhang, Yiqun Hui, Dewei Leng, Yanan Niu, Yang Song, and Kun Gai. 2023. PEPNet: Parameter and Embedding Personalized Network for Infusing with Personalized Prior Information. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD 2023, Long Beach, CA, USA, August 6-10, 2023, Ambuj K. Singh, Yizhou...

    work page doi:10.1145/3580305.3599884 2023

  3. [3]

    Zichuan Fu, Xiangyang Li, Chuhan Wu, Yichao Wang, Kuicai Dong, Xiangyu Zhao, Mengchen Zhao, Huifeng Guo, and Ruiming Tang. 2023. A Unified Frame- work for Multi-Domain CTR Prediction via Large Language Models.CoRR abs/2312.10743 (2023). doi:10.48550/ARXIV.2312.10743 arXiv:2312.10743

    work page doi:10.48550/arxiv.2312.10743 2023

  4. [4]

    Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. 2017. DeepFM: A Factorization-Machine based Neural Network for CTR Prediction. InProceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, IJCAI 2017, Melbourne, Australia, August 19-25, 2017, Carles Sierra (Ed.). ijcai.org, 1725–1731. doi:10.24963/IJCAI.2017/239

    work page doi:10.24963/ijcai.2017/239 2017

  5. [5]

    Hamilton, Zhitao Ying, and Jure Leskovec

    William L. Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive Represen- tation Learning on Large Graphs. InAdvances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, December 4-9, 2017, Long Beach, CA, USA, Isabelle Guyon, Ulrike von Luxburg, Samy Bengio, Hanna M. Wallach, Rob Fergus, S. V...

    work page 2017

  6. [6]

    Zhongxuan Han, Xiaolin Zheng, Chaochao Chen, Wenjie Cheng, and Yang Yao. 2023. Intra and Inter Domain HyperGraph Convolutional Network for Cross-Domain Recommendation. InProceedings of the ACM Web Conference 2023, WWW 2023, Austin, TX, USA, 30 April 2023 - 4 May 2023, Ying Ding, Jie Tang, Juan F. Sequeda, Lora Aroyo, Carlos Castillo, and Geert-Jan Houben ...

    work page doi:10.1145/3543507.3583402 2023

  7. [7]

    Peng Hao, Guangquan Zhang, Luis Martínez-López, and Jie Lu. 2019. Regularizing Knowledge Transfer in Recommendation With Tag-Inferred Correlation.IEEE Trans. Cybern.49, 1 (2019), 83–96. doi:10.1109/TCYB.2017.2764918

    work page doi:10.1109/tcyb.2017.2764918 2019

  8. [8]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yong-Dong Zhang, and Meng Wang. 2020. LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation. InProceedings of the 43rd International ACM SIGIR con- ference on research and development in Information Retrieval, SIGIR 2020, Virtual Event, China, July 25-30, 2020, Jimmy X. Huang, Yi Chang,...

    work page doi:10.1145/3397271.3401063 2020

  9. [9]

    McAuley, and Wayne Xin Zhao

    Yupeng Hou, Zhankui He, Julian J. McAuley, and Wayne Xin Zhao. 2023. Learn- ing Vector-Quantized Item Representation for Transferable Sequential Recom- menders. InProceedings of the ACM Web Conference 2023, WWW 2023, Austin, TX, USA, 30 April 2023 - 4 May 2023, Ying Ding, Jie Tang, Juan F. Sequeda, Lora Aroyo, Carlos Castillo, and Geert-Jan Houben (Eds.)....

    work page doi:10.1145/3543507.3583434 2023

  10. [10]

    Yupeng Hou, Shanlei Mu, Wayne Xin Zhao, Yaliang Li, Bolin Ding, and Ji-Rong Wen. 2022. Towards Universal Sequence Representation Learning for Recom- mender Systems. InKDD ’22: The 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Washington, DC, USA, August 14 - 18, 2022, Aidong Zhang and Huzefa Rangwala (Eds.). ACM, 585–593. doi:10.1145/...

    work page doi:10.1145/3534678.3539381 2022

  11. [11]

    Jingwen Hu, Yuchen Liu, Jinming Zhao, and Qin Jin. 2021. MMGCN: Multimodal Fusion via Deep Graph Convolution Network for Emotion Recognition in Conver- sation. InProceedings of the 59th Annual Meeting of the Association for Computa- tional Linguistics and the 11th International Joint Conference on Natural Language Processing, ACL/IJCNLP 2021, (Volume 1: L...

    work page doi:10.18653/v1/2021.acl-long.440 2021

  12. [12]

    Ziniu Hu, Yuxiao Dong, Kuansan Wang, Kai-Wei Chang, and Yizhou Sun. 2020. GPT-GNN: Generative Pre-Training of Graph Neural Networks. InKDD ’20: The 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Virtual Event, CA, USA, August 23-27, 2020, Rajesh Gupta, Yan Liu, Jiliang Tang, and B. Aditya Prakash (Eds.). ACM, 1857–1867. doi:10.1145/339...

    work page doi:10.1145/3394486.3403237 2020

  13. [13]

    Zepeng Huai, Yuji Yang, Mengdi Zhang, Zhongyi Zhang, Yichun Li, and Wei Wu. 2023. M2GNN: Metapath and Multi-interest Aggregated Graph Neural Network for Tag-based Cross-domain Recommendation. InProceedings of the 46th International ACM SIGIR Conference on Research and Development in Infor- mation Retrieval, SIGIR 2023, Taipei, Taiwan, July 23-27, 2023, Hs...

    work page doi:10.1145/3539618.3591720 2023

  14. [14]

    Semi-Supervised Classification with Graph Convolutional Networks

    Thomas N. Kipf and Max Welling. 2016. Semi-Supervised Classification with Graph Convolutional Networks.CoRRabs/1609.02907 (2016). arXiv:1609.02907 http://arxiv.org/abs/1609.02907

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  15. [15]

    Yakun Li, Lei Hou, and Juanzi Li. 2024. Towards Knowledge-Aware and Deep Re- inforced Cross-Domain Recommendation Over Collaborative Knowledge Graph. IEEE Trans. Knowl. Data Eng.36, 11 (2024), 7171–7187. doi:10.1109/TKDE.2024. 3391268

    work page doi:10.1109/tkde.2024 2024

  16. [16]

    Meng Liu, Jianjun Li, Guohui Li, and Peng Pan. 2020. Cross Domain Recommen- dation via Bi-directional Transfer Graph Collaborative Filtering Networks. In CIKM ’20: The 29th ACM International Conference on Information and Knowledge Management, Virtual Event, Ireland, October 19-23, 2020, Mathieu d’Aquin, Stefan Dietze, Claudia Hauff, Edward Curry, and Phil...

    work page doi:10.1145/3340531.3412012 2020

  17. [17]

    Huishi Luo, Yiwen Chen, Yiqing Wu, Fuzhen Zhuang, and Deqing Wang. 2024. One for Dozens: Adaptive REcommendation for All Domains with Counterfactual Augmentation.CoRRabs/2412.11905 (2024). doi:10.48550/ARXIV.2412.11905 arXiv:2412.11905

    work page doi:10.48550/arxiv.2412.11905 2024

  18. [18]

    Jiaqi Ma, Zhe Zhao, Xinyang Yi, Jilin Chen, Lichan Hong, and Ed H. Chi. 2018. Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture- of-Experts. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2018, London, UK, August 19-23, 2018, Yike Guo and Faisal Farooq (Eds.). ACM, 1930–1939...

    work page doi:10.1145/3219819.3220007 2018

  19. [19]

    Tong Man, Huawei Shen, Xiaolong Jin, and Xueqi Cheng. 2017. Cross-Domain Recommendation: An Embedding and Mapping Approach. InProceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, IJCAI 2017, Melbourne, Australia, August 19-25, 2017, Carles Sierra (Ed.). ijcai.org, 2464–2470. doi:10.24963/IJCAI.2017/343

    work page doi:10.24963/ijcai.2017/343 2017

  20. [20]

    Jianmo Ni, Jiacheng Li, and Julian J. McAuley. 2019. Justifying Recommendations using Distantly-Labeled Reviews and Fine-Grained Aspects. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, EMNLP-IJCNLP 2019, Hong Kong, China, November 3-7, 2019...

    work page doi:10.18653/v1/d19-1018 2019

  21. [21]

    Wentao Ning, Reynold Cheng, Jiajun Shen, Nur Al Hasan Haldar, Ben Kao, Xiao Yan, Nan Huo, Wai Kit Lam, Tian Li, and Bo Tang. 2022. Automatic Meta-Path Discovery for Effective Graph-Based Recommendation. InProceedings of the 31st ACM International Conference on Information & Knowledge Management, Atlanta, GA, USA, October 17-21, 2022, Mohammad Al Hasan and...

    work page doi:10.1145/3511808.3557244 2022

  22. [22]

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

    work page

  23. [23]

    Multi-domain Recommendation with Embedding Disentangling and Do- main Alignment. InProceedings of the 32nd ACM International Conference on Information and Knowledge Management, CIKM 2023, Birmingham, United King- dom, October 21-25, 2023, Ingo Frommholz, Frank Hopfgartner, Mark Lee, Michael Oakes, Mounia Lalmas, Min Zhang, and Rodrygo L. T. Santos (Eds.)....

    work page 2023

  24. [24]

    doi:10.1145/3583780.3614977

    work page doi:10.1145/3583780.3614977

  25. [25]

    Yi Ouyang, Bin Guo, Qianru Wang, and Zhiwen Yu. 2021. Cross-Domain Recom- mendation with Cross-Graph Knowledge Transfer Network. InICC 2021 - IEEE International Conference on Communications, Montreal, QC, Canada, June 14-23,

    work page 2021

  26. [26]

    2021 , journal =

    IEEE, 1–6. doi:10.1109/ICC42927.2021.9500882 Y. Chen et al

    work page doi:10.1109/icc42927.2021.9500882 2021

  27. [27]

    Jiezhong Qiu, Qibin Chen, Yuxiao Dong, Jing Zhang, Hongxia Yang, Ming Ding, Kuansan Wang, and Jie Tang. 2020. GCC: Graph Contrastive Coding for Graph Neural Network Pre-Training. InKDD ’20: The 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Virtual Event, CA, USA, August 23-27, 2020, Rajesh Gupta, Yan Liu, Jiliang Tang, and B. Aditya P...

    work page doi:10.1145/3394486.3403168 2020

  28. [28]

    Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt- Thieme. 2009. BPR: Bayesian Personalized Ranking from Implicit Feedback. InUAI 2009, Proceedings of the Twenty-Fifth Conference on Uncertainty in Ar- tificial Intelligence, Montreal, QC, Canada, June 18-21, 2009, Jeff A. Bilmes and Andrew Y. Ng (Eds.). AUAI Press, 452–461. https://ww...

    work page 2009

  29. [29]

    Leheng Sheng, An Zhang, Yi Zhang, Yuxin Chen, Xiang Wang, and Tat-Seng Chua. 2024. Language Models Encode Collaborative Signals in Recommendation. arXiv preprint arXiv:2407.05441(2024)

    work page arXiv 2024

  30. [30]

    Xiang-Rong Sheng, Liqin Zhao, Guorui Zhou, Xinyao Ding, Binding Dai, Qiang Luo, Siran Yang, Jingshan Lv, Chi Zhang, Hongbo Deng, and Xiaoqiang Zhu

    work page

  31. [31]

    One Model to Serve All: Star Topology Adaptive Recommender for Multi- Domain CTR Prediction. InCIKM ’21: The 30th ACM International Conference on Information and Knowledge Management, Virtual Event, Queensland, Australia, November 1 - 5, 2021, Gianluca Demartini, Guido Zuccon, J. Shane Culpepper, Zi Huang, and Hanghang Tong (Eds.). ACM, 4104–4113. doi:10....

    work page doi:10.1145/3459637 2021

  32. [32]

    Weiping Song, Chence Shi, Zhiping Xiao, Zhijian Duan, Yewen Xu, Ming Zhang, and Jian Tang. 2019. AutoInt: Automatic Feature Interaction Learning via Self- Attentive Neural Networks. InProceedings of the 28th ACM International Con- ference on Information and Knowledge Management, CIKM 2019, Beijing, China, November 3-7, 2019, Wenwu Zhu, Dacheng Tao, Xueqi ...

    work page doi:10.1145/3357384.3357925 2019

  33. [33]

    Mingchen Sun, Kaixiong Zhou, Xin He, Ying Wang, and Xin Wang. 2022. GPPT: Graph Pre-training and Prompt Tuning to Generalize Graph Neural Networks. InKDD ’22: The 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Washington, DC, USA, August 14 - 18, 2022, Aidong Zhang and Huzefa Rangwala (Eds.). ACM, 1717–1727. doi:10.1145/3534678.3539249

    work page doi:10.1145/3534678.3539249 2022

  34. [34]

    Hongyan Tang, Junning Liu, Ming Zhao, and Xudong Gong. 2020. Progressive Layered Extraction (PLE): A Novel Multi-Task Learning (MTL) Model for Person- alized Recommendations. InRecSys 2020: Fourteenth ACM Conference on Recom- mender Systems, Virtual Event, Brazil, September 22-26, 2020, Rodrygo L. T. Santos, Leandro Balby Marinho, Elizabeth M. Daly, Li Ch...

    work page doi:10.1145/3383313.3412236 2020

  35. [35]

    Petar Velickovic, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. 2018. Graph Attention Networks. In6th Interna- tional Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings. OpenReview.net. https: //openreview.net/forum?id=rJXMpikCZ

    work page 2018

  36. [36]

    Chen Wang, Yueqing Liang, Zhiwei Liu, Tao Zhang, and S Yu Philip. 2021. Pre- training graph neural network for cross domain recommendation. In2021 IEEE Third International Conference on Cognitive Machine Intelligence (CogMI). IEEE, 140–145

    work page 2021

  37. [37]

    Ruoxi Wang, Bin Fu, Gang Fu, and Mingliang Wang. 2017. Deep & Cross Network for Ad Click Predictions. InProceedings of the ADKDD’17, Halifax, NS, Canada, August 13 - 17, 2017. ACM, 12:1–12:7. doi:10.1145/3124749.3124754

    work page doi:10.1145/3124749.3124754 2017

  38. [38]

    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, SIGIR 2019, Paris, France, July 21-25, 2019, Benjamin Piwowarski, Max Chevalier, Éric Gaussier, Yoelle Maarek, Jian-Yun Nie, and Fal...

    work page doi:10.1145/3331184.3331267 2019

  39. [39]

    Wei Wei, Xubin Ren, Jiabin Tang, Qinyong Wang, Lixin Su, Suqi Cheng, Junfeng Wang, Dawei Yin, and Chao Huang. 2024. LLMRec: Large Language Models with Graph Augmentation for Recommendation. InProceedings of the 17th ACM Inter- national Conference on Web Search and Data Mining, WSDM 2024, Merida, Mexico, March 4-8, 2024, Luz Angelica Caudillo-Mata, Silvio ...

    work page doi:10.1145/3616855.3635853 2024

  40. [40]

    Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and Philip S. Yu. 2021. A Comprehensive Survey on Graph Neural Networks.IEEE Trans. Neural Networks Learn. Syst.32, 1 (2021), 4–24. doi:10.1109/TNNLS.2020. 2978386

    work page doi:10.1109/tnnls.2020 2021

  41. [41]

    Hamilton, and Jure Leskovec

    Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L. Hamilton, and Jure Leskovec. 2018. Graph Convolutional Neural Networks for Web-Scale Recommender Systems. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2018, London, UK, August 19-23, 2018, Yike Guo and Faisal Farooq (Eds.). ACM, 974...

    work page arXiv 2018

  42. [42]

    Junliang Yu, Xin Xia, Tong Chen, Lizhen Cui, Nguyen Quoc Viet Hung, and Hongzhi Yin. 2024. XSimGCL: Towards Extremely Simple Graph Contrastive Learning for Recommendation.IEEE Trans. Knowl. Data Eng.36, 2 (2024), 913–926. doi:10.1109/TKDE.2023.3288135

    work page doi:10.1109/tkde.2023.3288135 2024

  43. [43]

    Jinghao Zhang, Yanqiao Zhu, Qiang Liu, Shu Wu, Shuhui Wang, and Liang Wang

    work page

  44. [44]

    InMM ’21: ACM Multimedia Conference, Virtual Event, China, October 20 - 24, 2021, Heng Tao Shen, Yueting Zhuang, John R

    Mining Latent Structures for Multimedia Recommendation. InMM ’21: ACM Multimedia Conference, Virtual Event, China, October 20 - 24, 2021, Heng Tao Shen, Yueting Zhuang, John R. Smith, Yang Yang, Pablo César, Florian Metze, and Balakrishnan Prabhakaran (Eds.). ACM, 3872–3880. doi:10.1145/3474085.3475259

    work page doi:10.1145/3474085.3475259 2021

  45. [45]

    Qian Zhang, Jie Lu, Dianshuang Wu, and Guangquan Zhang. 2018. Cross- domain Recommendation with Consistent Knowledge Transfer by Subspace Alignment. InWeb Information Systems Engineering - WISE 2018 - 19th Inter- national Conference, Dubai, United Arab Emirates, November 12-15, 2018, Pro- ceedings, Part II (Lecture Notes in Computer Science, Vol. 11234), ...

    work page doi:10.1007/978-3-030-02925-8_5 2018

  46. [46]

    Qian Zhang, Jie Lu, Dianshuang Wu, and Guangquan Zhang. 2019. A Cross- Domain Recommender System With Kernel-Induced Knowledge Transfer for Overlapping Entities.IEEE Trans. Neural Networks Learn. Syst.30, 7 (2019), 1998–2012. doi:10.1109/TNNLS.2018.2875144

    work page doi:10.1109/tnnls.2018.2875144 2019

  47. [47]

    Yiyuan Zheng, Beibei Li, Beihong Jin, Rui Zhao, Weijiang Lai, and Tao Xiang

    work page

  48. [48]

    InMachine Learning and Knowledge Discovery in Databases

    Multi-intent Driven Contrastive Sequential Recommendation. InMachine Learning and Knowledge Discovery in Databases. Applied Data Science Track - European Conference, ECML PKDD 2024, Vilnius, Lithuania, September 9-13, 2024, Proceedings, Part IX (Lecture Notes in Computer Science, Vol. 14949), Albert Bifet, Tomas Krilavicius, Ioanna Miliou, and Slawomir No...

    work page doi:10.1007/978-3-031-70378-2_9 2024

  49. [49]

    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 Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI 2020, Christian Bessiere (Ed.). ijcai.org, 3001–3008. doi:10.24963/IJCAI.2020/415

    work page doi:10.24963/ijcai.2020/415 2020

  50. [50]

    ”] 4 Visit Amazon's Khardine Gray Page 6,277,419 in Books Recent Reviews ['I just finished a wonderful

    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. InWSDM ’22: The Fifteenth ACM International Conference on Web Search and Data Mining, Virtual Event / Tempe, AZ, USA, February 21 - 25, 2022, K. Selcuk Candan, Huan Liu, Leman ...

    work page doi:10.1145/3488560.3498392 2022

  51. [51]

    Specifically, we used the Faiss library to compute the top-20 nearest neighbors for each item/user based on cosine similarity

    Selection of 𝛾 Based on Similarity Distribution.To ensure that the threshold 𝛾 is both effective and efficient, we conducted a thor- ough analysis of the similarity distribution among the textual em- beddings of items and users across different domains. Specifically, we used the Faiss library to compute the top-20 nearest neighbors for each item/user base...

    work page

  52. [52]

    We also explored the impact of using lower similarity thresholds (e.g., [0.6, 0.7]) to under- stand the trade-offs between noise reduction and recall

    Extended Experiments with Different 𝛾 Values.To further in- vestigate the impact of 𝛾 on model performance, we conducted experiments with a wider range of 𝛾 values. We also explored the impact of using lower similarity thresholds (e.g., [0.6, 0.7]) to under- stand the trade-offs between noise reduction and recall. The results are summarized below: Analysi...

    work page

  53. [53]

    Real-World Data Analysis.We used the 8D subset from a real- world Amazon business scenario, which inherently contains spar- sity and noise. The statistics of missing data are as follows: Despite the sparsity and noise, our model achieved an AUC of 0.7561 and Recall@10 of 0.3382, demonstrating its robustness in real-world scenarios

    work page

  54. [54]

    Masking Simulation Experiments.To further assess robustness, we designed a series of simulation experiments by progressively masking different types of information. The types of information masked include: •ID information (type0), •reviews (type1), •titles (type2), •descriptions, features (type3), • numerical information like price, brand, and salesRank (...

    work page

  55. [55]

    • ID Information: Removing user and item IDs (type0) has minimal impact on performance, indicating that our model is not overly reliant on ID information

    Analysis. • ID Information: Removing user and item IDs (type0) has minimal impact on performance, indicating that our model is not overly reliant on ID information. • Review Information: Masking reviews (type1) leads to a more noticeable drop in performance, especially when 50% of reviews are removed. However, even without any review information, our mode...

    work page

  56. [56]

    The results are as follows:

    Cold-Start Experiment.We simulated a cold-start scenario by retaining only 5% of the original training data in the target domain (Sports) while keeping other data unchanged. The results are as follows:

    work page

  57. [57]

    • Sparse Data: The low performance of most models is pri- marily due to the extreme sparsity of training data in the target domain

    Analysis. • Sparse Data: The low performance of most models is pri- marily due to the extreme sparsity of training data in the target domain. This makes it challenging for models to generalize and perform well with minimal training data. • Cold-Start Capability: Our model significantly outper- forms other baselines, demonstrating its ability to adapt to c...

    work page