Pith. sign in

REVIEW 5 major objections 6 minor 80 references

Hybrid-Hierarchical Fashion Graph Attention Network for Compatibility-Oriented and Personalized Outfit Recommendation

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read FGAT jointly models outfit compatibility and personalized taste on a three-tier fashion graph, and reports gains over strong baselines on the POG dataset.

desk verdict A real but incremental extension of HFGN that overclaims on NDCG; the architecture and compatibility results are solid, but Table IV undercuts the ranking claim. read the letter →

arxiv 2508.11105 v2 pith:7Q6FMYIP submitted 2025-08-14 cs.LG cs.IR

classification cs.LGcs.IR
keywords graphattentionnetworksfashionrecommendationoutfitcompatibilitypersonalizationmultimodallearninghierarchicalcategoryco-occurrencePOGdataset
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 introduces FGAT, a graph attention network that treats fashion recommendation as one problem rather than two: it learns outfit compatibility and user preference together in a single three-tier graph of users, outfits, and items. The claim is that adding textual item descriptions alongside visual features, and weighting item-to-item messages by how often their categories co-occur in real outfits, gives better embeddings than prior hierarchical models. On the POG dataset, the authors report that FGAT beats the strong baseline HFGN on HR@10, Precision@10, Recall@10, NDCG@10, compatibility accuracy, and AUC. A sympathetic reader would care because e-commerce fashion recommendations are only useful when a suggestion is both coherent as an outfit and right for the shopper, and most prior systems optimize one side only.

What carries the argument

The load-bearing mechanism is self-attention message passing over the three-level user-outfit-item graph, with the item-to-item attention matrix initialized by category co-occurrence weights computed from training outfits. The transformation matrix in the attention coefficient is initialized with these weights, so items from categories that frequently appear together start with stronger compatibility signals before learning begins. Propagation proceeds item-to-item, then item-to-outfit, then outfit-to-user, with multi-head attention and element-wise products that let compatible neighbors contribute more, and a multi-view attention module scores each item's importance and compatibility across several semantic views to produce the final outfit compatibility score.

What would settle it

Take FGAT and replace the category co-occurrence initialization of the item-to-item attention weights with random or uniform weights, keeping everything else fixed; if the reported gains over HFGN survive, the co-occurrence prior is not the driver. A second check: hold out all outfit pairs containing an unseen category combination and ask whether FGAT still ranks the true complementary item above the three distractors in the fill-in-the-blank task; if accuracy collapses on those pairs, the model is memorizing co-occurrence statistics rather than learning compatibility.

Watch

Extended reading notes

Core claim

The central discovery is that a hierarchical fashion graph with attention can carry both compatibility and personalization at once, instead of treating them as separate tasks. Item embeddings start from pre-trained visual features and pre-trained Chinese textual features projected into a common 64-dimensional space; an item-to-item attention layer, initialized from category co-occurrence counts, propagates compatibility; item-to-outfit and outfit-to-user attention layers fold style and history into outfit and user embeddings; and a multi-view compatibility scorer reads the resulting item embeddings to judge whole outfits. The authors report HR@10 of 0.4286 versus HFGN's 0.2833, Precision@10 of 0.4424 versus 0.3390, NDCG@10 of 0.1340 versus 0.1241, and compatibility accuracy of 0.8956 versus 0.8797 on the POG dataset.

Load-bearing premise

The model assumes that how often two clothing categories appear together in the training outfits is a reliable signal of whether those categories, and the specific items inside them, are compatible, and that initializing the attention mechanism with these counts helps rather than biases the learned embeddings.

Editorial extensions

If this is right

  • Joint training on both losses lets one set of embeddings serve both outfit compatibility and personalized ranking, so improving one task need not come at the expense of the other.
  • Textual features from item titles carry complementary signals such as material, gender, and occasion, so adding them should help most on categories where images alone are ambiguous.
  • Category co-occurrence priors give the attention mechanism a sensible starting point, which should matter most in small-data regimes and for cold-start items whose categories are well represented.
  • Because the framework currently propagates only first-order paths, extending it to higher-order user-outfit-item paths is a direct next step that the paper leaves open.

Reading between the lines

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

  • Editorial extension: if the co-occurrence prior is the main source of gain, the same category-level weight matrix could be reused as a transfer signal for entirely new items, since the prior is computed over categories rather than individual item embeddings.
  • Editorial extension: the multi-view attention module is effectively a small interpretable component, and its learned view weights could be read out to tell users why an outfit matches, which the paper does not explore.
  • Editorial extension: the large HR@10 gap over HFGN combined with a modest NDCG@10 gap suggests the improvement is concentrated in whether the right outfit appears in the top ten, not in fine-grained ranking order; a position-aware analysis would test that reading.
  • Editorial extension: a natural stress test is to ablate the textual branch only; if the pre-trained textual features account for most of the gain, future work could concentrate on stronger text encoders rather than on graph depth.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes FGAT, a hierarchical three-level graph attention network for fashion outfit recommendation on the POG dataset. The model propagates information over user–outfit–item graphs, initializes item-item attention from category co-occurrence statistics, fuses ResNet visual features with BERT textual features, and jointly optimizes BPR losses for personalized outfit recommendation and outfit compatibility scoring. The authors report gains over HFGN across HR@10, Recall@10, Precision@10, NDCG@10, Accuracy, and AUC, and claim state-of-the-art performance.

Significance. If validated, the architecture would be a plausible contribution: it unifies compatibility and personalization in one end-to-end framework, uses a legitimate category co-occurrence prior (computed from training data, not circular), and ships a concrete evaluation on a real-world dataset (POG). The multimodal fusion and R-view compatibility scoring are reasonable design choices. However, the paper's load-bearing claim of outperforming state-of-the-art baselines is contradicted by its own Table IV on NDCG@10, and the evaluation protocol has inconsistencies and lacks statistical controls. As written, the central claim is not supported by the evidence.

major comments (5)
  1. [Abstract, Section I, Section IV.F, Table IV] The claim that FGAT 'outperforms strong baselines such as HFGN, achieving notable improvements in ... NDCG' is internally contradicted by Table IV. FGAT's NDCG@10 is 0.1340, while Hg-PDC achieves 0.3532, LightGCN 0.2882, BPR 0.2633, and BCDSVD++ 0.1666; FGAT only beats HFGN (0.1241), by 0.0099. The abstract and introduction overstate the NDCG result, and Section IV.F itself limits the NDCG claim to an improvement over HFGN. This is a load-bearing inconsistency in the headline result.
  2. [Section III.A and Section IV.B] The data split is described inconsistently. Section III.A states '80% of each user’s interactions are used for training, 10% of that training set is used for validation, and the remaining 20% is used for testing', which is arithmetically ambiguous (10% of 80% is 8%, leaving 12%, not 20%). Section IV.B states '80% for training, 10% for validation, and 10% for testing'. The precise protocol must be stated unambiguously, including how the 1,647 test outfits and 3,126 test items in Table II are selected relative to the recommendation-task split, because the reported metrics are not reproducible otherwise.
  3. [Section IV.F, Table IV] The comparison against baselines is not demonstrably controlled. Table IV has numerous missing entries (e.g., DTNM lacks HR@10, Try-On-CM lacks NDCG@10, Precision@10, and Recall@10, RankBPR only reports Recall@10), and the text refers to a 'last column' of Table IV that does not exist in the printed table. Without a common experimental protocol, identical negative sampling, and fair hyperparameter tuning for every baseline, the relative ranking in Table IV cannot be interpreted. The authors should either re-run all baselines under the exact same setup or explicitly report the source and conditions of each baseline number.
  4. [Section IV.E and Section IV.F, Tables IV–VI] No error bars, standard deviations, or significance tests are reported. Several claimed improvements are small in absolute terms (e.g., NDCG@10 from 0.1241 to 0.1340, Accuracy from 0.8797 to 0.8956, AUC from 0.875 to 0.8974), and with a single run it is impossible to distinguish genuine gains from optimization noise. The authors should report multiple seeds with mean and variance, and, where appropriate, paired significance tests.
  5. [Section IV.D and Section IV.H] The compatibility evaluation protocol is underspecified. The FLTB task description says one item is masked and three items are randomly selected from other outfits, but it does not specify whether the negative candidates are sampled from the same category, how many random seeds are used, or how Accuracy and AUC are computed from this procedure. The paper also claims accuracy is used 'when compatible and incompatible classes are balanced', but no class-balance analysis is provided for the FLTB test set.
minor comments (6)
  1. [Section III.C] Equations (1) through (5) appear as blank placeholders in the manuscript, so the visual and textual feature extraction formulas are missing.
  2. [Section V] The first two bullets in the future-work list are identical: 'Incorporating time-aware and dynamic user-item interactions...' is duplicated.
  3. [Section IV.F] The phrase 'suggesting more prices results' appears to be a typo; it should likely be 'precise results' or similar.
  4. [Section III.D] The attention weight symbol is rendered inconsistently as '∝i,j' in the text but should be the Greek alpha 'αi,j' consistent with Table III.
  5. [Table II and Section III.A] The text says '1,647 unused/unappeared items in training set are selected as negative samples for the test set', but Table II reports 1,647 test outfits and 3,126 test items; the relationship between these numbers should be clarified.
  6. [Section I and V] The claim that the model is 'efficient and scalable' is not supported by any complexity analysis or runtime comparison, especially given the stated scalability limitations in the Conclusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: FGAT's claimed predictions are learned on held-out data, and the category co-occurrence prior is a legitimate training-derived feature rather than a fitted target.

full rationale

The paper's derivation chain does not reduce to its own inputs. The item-item attention initialization in Eqs. (6)-(10) uses category co-occurrence counts computed from the training outfits; this is a legitimate feature/prior rather than a fitted label, and the reported HR@10, Precision@10, Recall@10, NDCG@10, AUC, and Accuracy are evaluated on a held-out test split (Sections III.A and IV.B), so no metric is forced by construction. The compatibility score in Eqs. (22)-(24) is trained with BPR against positive training outfits and randomly generated negatives, making it a learned function rather than a restatement of co-occurrence statistics. The two self-citations in the related-work section ([14], [31]) are contextual and are not load-bearing for the FGAT architecture or for its claimed improvements; no uniqueness theorem or ansatz is imported from the authors' own prior work. The internal inconsistency between the split descriptions in Section III.A and Section IV.B, and the fact that Table IV only shows NDCG improvement over HFGN while other baselines score higher on NDCG, are correctness and verifiability concerns rather than circularity concerns. No self-definitional, fitted-input-called-prediction, self-citation-reduction, or renaming step can be exhibited from the manuscript text.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The model relies on standard pretrained extractors and a hand-crafted co-occurrence prior. No new physical or conceptual entity is introduced.

free parameters (3)
  • Number of semantic views R = 6
    Chosen by hand for the R-view attention in Equation (22); no sensitivity analysis is provided.
  • Number of attention heads = 4
    Set in Section III.D; no ablation study is provided.
  • Embedding dimension d = 64
    Set in Section III.C; standard choice with no tuning rationale.
assumptions (4)
  • domain assumption Category co-occurrence is a reliable proxy for fashion compatibility
    The paper uses w(ci,cj) from Equation (6) to initialize and guide attention in item-item propagation.
  • domain assumption Pretrained ResNet-152 and Chinese BERT provide adequate visual and textual representations for fashion items
    The model relies on these extracted features without fine-tuning on the fashion domain.
  • domain assumption BPR ranking loss accurately captures user preference and outfit compatibility
    Equations (19) and (20) use BPR for both tasks.
  • ad hoc to paper First-order graph propagation is sufficient to capture user preferences and compatibility
    The paper explicitly defers higher-order paths to future work (Section III).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hybrid-Hierarchical Fashion Graph Attention Network for Compatibility-Oriented and Personalized Outfit Recommendation." pith.science (2026). https://pith.science/paper/7Q6FMYIP

@misc{pith2026250811105,
  author       = {Pith},
  title        = {Pith review of: Hybrid-Hierarchical Fashion Graph Attention Network for Compatibility-Oriented and Personalized Outfit Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7Q6FMYIP}},
  note         = {Machine review of arXiv:2508.11105}
}
read the original abstract

The rapid expansion of the fashion industry and the growing variety of products have made it increasingly challenging for users to identify compatible items on e-commerce platforms. Effective fashion recommendation systems are therefore crucial for filtering irrelevant options and suggesting suitable ones. However, simultaneously addressing outfit compatibility and personalized recommendations remains a significant challenge, as these aspects are typically treated independently in existing studies, thereby overlooking the complex interactions between items and user preferences. This research introduces a new framework named FGAT, which leverages a hierarchical graph representation together with graph attention mechanisms to address this problem. The framework constructs a three-tier graph of users, outfits, and items, integrating visual and textual features to jointly model outfit compatibility and user preferences. By dynamically weighting node importance during representation propagation, the graph attention mechanism captures key interactions and produces precise embeddings for both user preferences and outfit compatibility. Evaluated on the POG dataset, FGAT outperforms strong baselines such as HFGN, achieving notable improvements in accuracy, precision, HR, recall, and NDCG. These results demonstrate that combining multimodal visual and textual features with a hierarchical graph structure and attention mechanisms significantly enhances the effectiveness and efficiency of personalized fashion recommendation systems.

Figures

Figures reproduced from arXiv: 2508.11105 by the authors.

Figure 1
Figure 1. A set of items forms a compatible outfit. In this study, we introduce FGAT (Fashion Graph Attention Network), a novel Hybrid-Hierarchical Fashion recommendation framework designed to jointly address the dual challenges of outfit compatibility and user preference personalization. Unlike prior approaches that treat these aspects independently, our model integrates them into a unified, end-to-end framework. The motivat… view at source ↗
Figure 2
Figure 2. illustrates the general flow of our proposed method. As shown, the framework consists of three main components: • We initialize embeddings for users and outfits using IDs, and for items using visual-textual features. • Updating embeddings via GATs using first-order paths. It should be noted that in this method we use first-order paths (outfit–item and user–outfit), while higher-order paths (such as user–outfit–item)… view at source ↗
Figure 3
Figure 3. A three-level fashion graph of users, outfits, and items. TABLE III. SUMMARY OF THE MAIN NOTATIONS. Notation Explanation u A user in the system. o An outfit in the system. i, j Indices representing items in outfits. ci, cj The category of item 𝑖 and item 𝑗, respectively. R d Embedding d-dimensional real vector space. xᵥ(i) Visual features of item i. xt(i) Textual features of item i. e ̂v(i) Visual embedding of item … view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: shows the initial embedding of the user, outfit, and item nodes. D. propagating information and updating embeddings via self-attention mechanism Recent research in GNNs ([3],[5],[10],[11],[55],[67], [68]) has shown that propagating information in graph structures can e…
Figure 5
Figure 5. Figure 5: , where the nodes, categories, and edges represent the weighted co-occurrence of categories in the outfits. Any two categories that appear together more frequently have a stronger relationship (i.e., higher weight) in this graph. For example, [PITH_FULL_IMAGE:figures/…
Figure 6
Figure 6. Figure 6: Top 5 pairs of categories that co-occur most frequently in the outfits [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 9
Figure 9. Figure 9: Visualization of a three-level fashion graph using the Gephi application: a) Outfits purchased by a large number of users are represented with many edges. b) Items that appear in multiple outfits are also represented with many edges [PITH_FULL_IMAGE:figures/full_fig_p…
Figure 10
Figure 10. Figure 10: Heatmap and scatter diagram of category co-occurrence in the same outfits [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 11
Figure 11. Figure 11: An outfit showing different attention weights for a pair of items. The self-attention mechanism in the outfit items subgraph stage adjusts attention weights based on co-occurrence relationships, such that more related items receive more weight. Category co-occurrence …
Figure 12
Figure 12. Figure 12: Evaluation charts of various metrics during model training over 100 epochs [PITH_FULL_IMAGE:figures/full_fig_p009_12.png]
Figure 13
Figure 13. Figure 13: Three-level hierarchical graph for recommending the top 5 outfits to user 25 - Link prediction [PITH_FULL_IMAGE:figures/full_fig_p012_13.png]
Figure 14
Figure 14. Figure 14: Previously purchased outfits of user 25 and the model’s recommended outfits [PITH_FULL_IMAGE:figures/full_fig_p012_14.png]
Figure 15
Figure 15. Figure 15: Example of the FLTB task. The model selects the item with the highest compatibility score as the complementary item for the outfit. TABLE VII. EXAMPLE OF AN R-VIEW ATTENTION MATRIX. EACH ROW REPRESENTS A VIEW, AND EACH COLUMN SHOWS THE IMPORTANCE OF AN ITEM IN THAT VI…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

80 extracted references · 38 canonical work pages

  1. [1]

    Recommendation Systems For E-commerce Systems An Overview,

    F. T. Abdul Hussien, A. M. S. Rahma, and H. B. Abdul Wahab, “Recommendation Systems For E-commerce Systems An Overview,” J. Phys. Conf. Ser., vol. 1897, no. 1, p. 012024, May 2021, doi: 10.1088/1742-6596/1897/1/012024

  2. [2]

    Learning Visual Clothing Style with Heterogeneous Dyadic Co - Occurrences,

    A. Veit, B. Kovacs, S. Bell, J. McAuley, K. Bala, and S. Belongie, “Learning Visual Clothing Style with Heterogeneous Dyadic Co - Occurrences,” in 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, Chile: IEEE, Dec. 2015, pp. 4642 –4650. doi: 10.1109/ICCV.2015.527

  3. [3]

    Multimodal Compatibility Modeling via Exploring the Consistent and Complementary Correlations,

    W. Guan, H. Wen, X. Song, C. -H. Yeh, X. Chang, and L. Nie, “Multimodal Compatibility Modeling via Exploring the Consistent and Complementary Correlations,” in Proceedings of the 29th ACM International Conference on Multimedia, Virtual Event China: ACM, Oct. 2021, pp. 2299–2307. doi: 10.1145/3474085.3475392

  4. [4]

    A decision tree based recommender system,

    A. Gershman, A. Meisels, K.-H. Lüke, L. Rokach, A. Schclar, and A. Sturm, “A decision tree based recommender system,” presented at the 10th International Conferenceon Innovative Internet Community Systems (I2CS) – Jubilee Edition 2010 –, Gesellschaft für Informatik e.V., 2010, pp. 170 –179. Accessed: Aug. 02, 2024. [Online]. Available: https://dl.gi.de/ha...

  5. [5]

    Semi -Supervised Classification with Graph Convolutional Networks,

    T. N. Kipf and M. Welling, “Semi -Supervised Classification with Graph Convolutional Networks,” Feb. 22, 2017, arXiv: arXiv:1609.02907. doi: 10.48550/arXiv.1609.02907

  6. [6]

    Inductive Representation Learning on Large Graphs,

    W. Hamilton, Z. Ying, and J. Leskovec, “Inductive Representation Learning on Large Graphs,” in Advances in Neural Information Processing Systems, Curran Associates, Inc., 2017. Accessed: Aug. 14 01, 2024. [Online]. Available: https://proceedings.neurips.cc/paper/2017/hash/5dd9db5e033da9c6fb 5ba83c7a7ebea9-Abstract.html

  7. [7]

    FashionViL: Fashion-Focused Vision-and-Language Representation Learning,

    X. Han, L. Yu, X. Zhu, L. Zhang, Y. -Z. Song, and T. Xiang, “FashionViL: Fashion-Focused Vision-and-Language Representation Learning,” in Computer Vision – ECCV 2022, S. Avidan, G. Brostow, M. Cissé, G. M. Farinella, and T. Hassner, Eds., Cham: Springer Nature Switzerland, 2022, pp. 634–651. doi: 10.1007/978-3- 031-19833-5_37

  8. [8]

    GraRep: Learning Graph Representations with Global Structural Information,

    S. Cao, W. Lu, and Q. Xu, “GraRep: Learning Graph Representations with Global Structural Information,” in Proceedings of the 24th ACM International on Conference on Information and Knowledge Management, Melbourne Australia: ACM, Oct. 2015, pp. 891 –900. doi: 10.1145/2806416.2806512

Show all 80 references
  1. [9]

    Hierarchical Neural Networks (HNN): Using TensorFlow to build HNN,

    R. Fontenot, J. Lazarus, P. Rudick, and A. Sgambellone, “Hierarchical Neural Networks (HNN): Using TensorFlow to build HNN,” SMU Data Sci. Rev., vol. 6, no. 2, Sep. 2022, [Online]. Available: https://scholar.smu.edu/datasciencereview/vol6/iss2/4

  2. [10]

    Hierarchical Fashion Graph Network for Personalized Outfit Recommendation,

    X. Li, X. Wang, X. He, L. Chen, J. Xiao, and T. -S. Chua, “Hierarchical Fashion Graph Network for Personalized Outfit Recommendation,” May 26, 2020, arXiv: arXiv:2005.12566. Accessed: Feb. 01, 2024. [Online]. Available: http://arxiv.org/abs/2005.12566

  3. [11]

    Dressing as a Whole: Outfit Compatibility Learning Based on Node -wise Graph Neural Networks,

    Z. Cui, Z. Li, S. Wu, X. -Y. Zhang, and L. Wang, “Dressing as a Whole: Outfit Compatibility Learning Based on Node -wise Graph Neural Networks,” in The World Wide Web Conference, in WWW ’19. New York, NY, USA: Association for Computing Machinery, May 2019, pp. 307–317. doi: 10...

  4. [12]

    Learning Binary Code for Personalized Fashion Recommendation,

    Z. Lu, Y. Hu, Y. Jiang, Y. Chen, and B. Zeng, “Learning Binary Code for Personalized Fashion Recommendation,” presented at the Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 10562 –10570. Accessed: Aug. 19,

  5. [13]

    Marcus, Graph Theory, vol

    D. Marcus, Graph Theory, vol. 53. in AMS/MAA Textbooks, vol. 53. Providence, Rhode Island: American Mathematical Society, 2020. doi: 10.1090/text/053

  6. [14]

    Sentiment -Driven Community Detection in a Network of Perfume Preferences,

    K. Kalashi, S. Saed, and B. Teimourpour, “Sentiment -Driven Community Detection in a Network of Perfume Preferences,” Oct. 28, 2024, arXiv: arXiv:2410.19177. doi: 10.48550/arXiv.2410.19177

  7. [15]

    Graph Representation Learning and Its Applications: A Survey,

    V. T. Hoang, H.-J. Jeon, E.-S. You, Y. Yoon, S. Jung, and O. -J. Lee, “Graph Representation Learning and Its Applications: A Survey,” Sensors, vol. 23, no. 8, Art. no. 8, Jan. 2023, doi: 10.3390/s23084168

  8. [16]

    DeepWalk: Online Learning of Social Representations,

    B. Perozzi, R. Al -Rfou, and S. Skiena, “DeepWalk: Online Learning of Social Representations,” in Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, Aug. 2014, pp. 701–710. doi: 10.1145/2623330.2623732

  9. [17]

    Asymmetric Transitivity Preserving Graph Embedding,

    M. Ou, P. Cui, J. Pei, Z. Zhang, and W. Zhu, “Asymmetric Transitivity Preserving Graph Embedding,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco California USA: ACM, Aug. 2016, pp. 1105–1114. doi: 10.1145/2...

  10. [18]

    GrAR: A novel framework for Graph Alignment based on Relativity concept,

    M. A. Soltanshahi, B. Teimourpour, T. Khatibi, and H. Zare, “GrAR: A novel framework for Graph Alignment based on Relativity concept,” Expert Syst. Appl., vol. 187, p. 115908, Jan. 2022, doi: 10.1016/j.eswa.2021.115908

  11. [19]

    A Comprehensive Survey on Graph Neural Networks,

    Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and P. S. Yu, “A Comprehensive Survey on Graph Neural Networks,” IEEE Trans. Neural Netw. Learn. Syst., vol. 32, no. 1, pp. 4 –24, Jan. 2021, doi: 10.1109/TNNLS.2020.2978386

  12. [20]

    Neural Graph Collaborative Filtering,

    X. Wang, X. He, M. Wang, F. Feng, and T. -S. Chua, “Neural Graph Collaborative Filtering,” in Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval, Jul. 2019, pp. 165 –174. doi: 10.1145/3331184.3331267

  13. [21]

    Graph Attention Networks,

    P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Liò, and Y. Bengio, “Graph Attention Networks,” Feb. 04, 2018, arXiv: arXiv:1710.10903. Accessed: Aug. 01, 2024. [Online]. Available: http://arxiv.org/abs/1710.10903

  14. [22]

    Attribute -aware heterogeneous graph network for fashion compatibility prediction,

    Z. Zhou, View Profile, Z. Su, View Profile, R. Wang, and View Profile, “Attribute -aware heterogeneous graph network for fashion compatibility prediction,” Neurocomputing, vol. 495, no. C, pp. 62 – 74, Jul. 2022, doi: 10.1016/j.neucom.2022.04.121

  15. [23]

    Fashion Recommendation Systems, Models and Methods: A Review,

    S. Chakraborty, Md. S. Hoque, N. Rahman Jeem, M. C. Biswas, D. Bardhan, and E. Lobaton, “Fashion Recommendation Systems, Models and Methods: A Review,” Informatics, vol. 8, no. 3, p. 49, Jul. 2021, doi: 10.3390/informatics8030049

  16. [25]

    Leveraging Two Types of Global Graph for Sequential Fashion Recommendation,

    Y. Ding, Y. Ma, W. K. Wong, and T. -S. Chua, “Leveraging Two Types of Global Graph for Sequential Fashion Recommendation,” in Proceedings of the 2021 International Conference on Multimedia Retrieval, Aug. 2021, pp. 73–81. doi: 10.1145/3460426.3463638

  17. [26]

    Deep Learning-Based Recommendation System: Systematic Review and Classification,

    C. Li, I. Ishak, H. Ibrahim, M. Zolkepli, F. Sidi, and C. Li, “Deep Learning-Based Recommendation System: Systematic Review and Classification,” IEEE Access, vol. 11, pp. 113790–113835, 2023, doi: 10.1109/ACCESS.2023.3323353

  18. [27]

    Outfit compatibility model using fully connected self-adjusting graph neural network,

    H. Liu, L. Li, N. Yu, K. Ma, T. Peng, and X. Hu, “Outfit compatibility model using fully connected self-adjusting graph neural network,” Vis. Comput., vol. 40, no. 11, pp. 8331 –8343, Nov. 2024, doi: 10.1007/s00371-023-03238-6

  19. [28]

    Graph representation learning: a survey,

    F. Chen, Y. -C. Wang, B. Wang, and C. -C. J. Kuo, “Graph representation learning: a survey,” APSIPA Trans. Signal Inf. Process., vol. 9, no. 1, 2020, doi: 10.1017/ATSIP.2020.13

  20. [29]

    Learning Color Compatibility in Fashion Outfits,

    H. Zhang, X. Yang, J. Tan, C. -H. Wu, J. Wang, and C. -C. J. Kuo, “Learning Color Compatibility in Fashion Outfits,” Jul. 05, 2020, arXiv: arXiv:2007.02388. doi: 10.48550/arXiv.2007.02388

  21. [30]

    Analysis of Clothing Image Classification Models: A Comparison Study between Traditional Machine Learning and Deep Learning Models,

    J. Xu, Y. Wei, A. Wang, H. Zhao, and D. Lefloch, “Analysis of Clothing Image Classification Models: A Comparison Study between Traditional Machine Learning and Deep Learning Models,” Fibres Text. East. Eur., vol. 30, no. 5, pp. 66 –78, Oct. 2022, doi: 10.2478/ftee-2022-0046

  22. [31]

    An Efficient Multiple Convolutional Neural Network Model (MCNN-14) for Fashion Image Classification,

    S. Saed, B. Teimourpour, K. Kalashi, and M. A. Soltanshahi, “An Efficient Multiple Convolutional Neural Network Model (MCNN-14) for Fashion Image Classification,” in 2024 10th International Conference on Web Research (ICWR), Apr. 2024, pp. 13 –21. doi: 10.1109/ICWR61162.2024.10533341

  23. [32]

    Fashion and Apparel Classification using Convolutional Neural Networks,

    A. Schindler, T. Lidy, S. Karner, and M. Hecker, “Fashion and Apparel Classification using Convolutional Neural Networks,” Nov. 11, 2018, arXiv: arXiv:1811.04374. doi: 10.48550/arXiv.1811.04374

  24. [33]

    ImageNet Classification with Deep Convolutional Neural Networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” in Advances in Neural Information Processing Systems, Curran Associates, Inc., 2012. Accessed: Aug. 17, 2024. [Online]. Available: https://proceedings.neurips.cc/p...

  25. [34]

    Deep Residual Learning for Image Recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” presented at the Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 770–778. Accessed: Aug. 17, 2024. [Online]. Available: https://openaccess.thecvf.com/c...

  26. [35]

    DeepFashion: Powering Robust Clothes Recognition and Retrieval With Rich Annotations,

    Z. Liu, P. Luo, S. Qiu, X. Wang, and X. Tang, “DeepFashion: Powering Robust Clothes Recognition and Retrieval With Rich Annotations,” presented at the Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 1096 –1104. Accessed: Aug. 17, 2024. ...

  27. [36]

    Personalized Fashion Recommendations for Diverse Body Shapes with Contrastive Multimodal Cross-Attention Network,

    MaJianghong, SunHuiyue, YangDezhao, and ZhangHaijun, “Personalized Fashion Recommendations for Diverse Body Shapes with Contrastive Multimodal Cross-Attention Network,” ACM Trans. Intell. Syst. Technol., Jul. 2024, doi: 10.1145/3637217

  28. [37]

    Heterogeneous -Grained Multi -Modal Graph Network for Outfit Recommendation,

    R. Xu, J. Wang, and Y. Li, “Heterogeneous -Grained Multi -Modal Graph Network for Outfit Recommendation,” IEEE Trans. Emerg. Top. Comput. Intell., vol. 8, no. 2, pp. 1788 –1799, Apr. 2024, doi: 10.1109/TETCI.2024.3358190

  29. [38]

    Study of AI - Driven Fashion Recommender Systems,

    S. Shirkhani, H. Mokayed, R. Saini, and H. Y. Chai, “Study of AI - Driven Fashion Recommender Systems,” SN Comput. Sci., vol. 4, no. 5, p. 514, Jul. 2023, doi: 10.1007/s42979-023-01932-9

  30. [39]

    Learning Type -Aware Embeddings for Fashion Compatibility,

    M. I. Vasileva, B. A. Plummer, K. Dusad, S. Rajpal, R. Kumar, and D. Forsyth, “Learning Type -Aware Embeddings for Fashion Compatibility,” presented at the Proceedings of the European Conference on Computer Vision (ECCV), 2018, pp. 390 –405. Accessed: Aug. 17, 2024. [Online]. ...

  31. [40]

    Dress Style Recommendation Based on Female Body Shapes,

    W. Zong, “Dress Style Recommendation Based on Female Body Shapes,” Aug. 2022, Accessed: Aug. 17, 2024. [Online]. Available: https://hdl.handle.net/1813/112171

  32. [41]

    Affective Image Content Analysis: Two Decades Review and New Perspectives,

    S. Zhao et al., “Affective Image Content Analysis: Two Decades Review and New Perspectives,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 10, pp. 6729 –6751, Oct. 2022, doi: 10.1109/TPAMI.2021.3094362

  33. [42]

    Recommender systems for sustainability: overview and research issues,

    A. Felfernig et al., “Recommender systems for sustainability: overview and research issues,” Front. Big Data, vol. 6, Oct. 2023, doi: 10.3389/fdata.2023.1284511

  34. [43]

    Term -weighting approaches in automatic text retrieval,

    G. Salton and C. Buckley, “Term -weighting approaches in automatic text retrieval,” Inf. Process. Manag., vol. 24, no. 5, pp. 513 –523, Jan. 1988, doi: 10.1016/0306-4573(88)90021-0

  35. [44]

    Efficient Estimation of Word Representations in Vector Space,

    T. Mikolov, K. Chen, G. Corrado, and J. Dean, “Efficient Estimation of Word Representations in Vector Space,” Sep. 06, 2013, arXiv: arXiv:1301.3781. doi: 10.48550/arXiv.1301.3781

  36. [45]

    Attention Is All You Need,

    A. Vaswani et al., “Attention Is All You Need,” Aug. 01, 2023, arXiv: arXiv:1706.03762. doi: 10.48550/arXiv.1706.03762

  37. [46]

    BERT: Pre - training of Deep Bidirectional Transformers for Language Understanding,

    J. Devlin, M. -W. Chang, K. Lee, and K. Toutanova, “BERT: Pre - training of Deep Bidirectional Transformers for Language Understanding,” May 24, 2019, arXiv: arXiv:1810.04805. doi: 10.48550/arXiv.1810.04805

  38. [47]

    A Survey of Vision -Language Pre-Trained Models,

    Y. Du, Z. Liu, J. Li, and W. X. Zhao, “A Survey of Vision -Language Pre-Trained Models,” Jul. 15, 2022, arXiv: arXiv:2202.10936. doi: 10.48550/arXiv.2202.10936

  39. [48]

    BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations from Transformer,

    F. Sun et al., “BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations from Transformer,” in Proceedings of the 28th ACM International Conference on Information and Knowledge Management, Beijing China: ACM, Nov. 2019, pp. 1441–1450. doi: 10.1145/3357384.3357895

  40. [49]

    Probabilistic Latent Semantic Analysis,

    T. Hofmann, “Probabilistic Latent Semantic Analysis,” Jan. 23, 2013, arXiv: arXiv:1301.6705. doi: 10.48550/arXiv.1301.6705

  41. [50]

    Neural Collaborative Filtering,

    X. He, L. Liao, H. Zhang, L. Nie, X. Hu, and T. -S. Chua, “Neural Collaborative Filtering,” in Proceedings of the 26th International Conference on World Wide Web, in WWW ’17. Republic and Canton of Geneva, CHE: International World Wide Web Conferences Steering Committee, Apr. ...

  42. [51]

    Context -Aware Visual Compatibility Prediction,

    G. Cucurull, P. Taslakian, and D. Vazquez, “Context -Aware Visual Compatibility Prediction,” presented at the Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 12617 –12626. Accessed: Aug. 19, 2024. [Online]. Available: https://openac...

  43. [52]

    Style2Vec: Representation Learning for Fashion Items from Style Sets,

    H. Lee, J. Seol, and S. Lee, “Style2Vec: Representation Learning for Fashion Items from Style Sets,” Aug. 14, 2017, arXiv: arXiv:1708.04014. doi: 10.48550/arXiv.1708.04014

  44. [53]

    OutfitTransformer: Learning Outfit Representations for Fashion Recommendation,

    R. Sarkar et al., “OutfitTransformer: Learning Outfit Representations for Fashion Recommendation,” in 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), Waikoloa, HI, USA: IEEE, Jan. 2023, pp. 3590 –3598. doi: 10.1109/WACV56688.2023.00359

  45. [54]

    POG: Personalized Outfit Generation for Fashion Recommendation at Alibaba iFashion,

    W. Chen et al., “POG: Personalized Outfit Generation for Fashion Recommendation at Alibaba iFashion,” in Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, in KDD ’19. New York, NY, USA: Association for Computing Machinery, Jul. 2...

  46. [55]

    Partially Supervised Compatibility Modeling,

    W. Guan et al., “Partially Supervised Compatibility Modeling,” IEEE Trans. Image Process., vol. 31, pp. 4733 –4745, 2022, doi: 10.1109/TIP.2022.3187290

  47. [56]

    N -GCN: Multi- scale Graph Convolution for Semi -supervised Node Classification,

    S. Abu-El-Haija, A. Kapoor, B. Perozzi, and J. Lee, “N -GCN: Multi- scale Graph Convolution for Semi -supervised Node Classification,” in Proceedings of The 35th Uncertainty in Artificial Intelligence Conference, PMLR, Aug. 2020, pp. 841 –851. Accessed: Aug. 19,

  48. [57]

    Graph Neural News Recommendation with Unsupervised Preference Disentanglement,

    L. Hu et al., “Graph Neural News Recommendation with Unsupervised Preference Disentanglement,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, D. Jurafsky, J. Chai, N. Schluter, and J. Tetreault, Eds., Online: Association for Computa...

  49. [58]

    Available: https://proceedings.mlr.press/v115/abu-el- haija20a.html

    [Online]. Available: https://proceedings.mlr.press/v115/abu-el- haija20a.html

  50. [59]

    Fashion Recommendation: Outfit Compatibility using GNN,

    S. Gulati, “Fashion Recommendation: Outfit Compatibility using GNN,” Apr. 28, 2024, arXiv: arXiv:2404.18040. doi: 10.48550/arXiv.2404.18040

  51. [60]

    Bi- directional Heterogeneous Graph Hashing towards Efficient Outfit Recommendation,

    W. Guan, X. Song, H. Zhang, M. Liu, C. -H. Yeh, and X. Chang, “Bi- directional Heterogeneous Graph Hashing towards Efficient Outfit Recommendation,” in Proceedings of the 30th ACM International Conference on Multimedia, in MM ’22. New York, NY, USA: Association for Computing M...

  52. [61]

    VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback,

    R. He and J. McAuley, “VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback,” Proc. AAAI Conf. Artif. Intell., vol. 30, no. 1, Art. no. 1, Feb. 2016, doi: 10.1609/aaai.v30i1.9973

  53. [62]

    Heterogeneous Graph Neural Network for Fashion Outfit Compatibility Prediction,

    Mingming L., Qingming G., Ya Z., and Xiankang Y., “Heterogeneous Graph Neural Network for Fashion Outfit Compatibility Prediction,” J. Comput. -Aided Des. Comput. Graph., vol. 36, no. 9, pp. 1351 – 1361, 2024, doi: 10.3724/SP.J.1089.2024.19974

  54. [63]

    Fashion Compatibility Modeling through a Multi -modal Try -on-guided Scheme,

    X. Dong, J. Wu, X. Song, H. Dai, and L. Nie, “Fashion Compatibility Modeling through a Multi -modal Try -on-guided Scheme,” in Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, in SIGIR ’20. New York, NY, USA: Asso...

  55. [64]

    PAI - BPR: Personalized Outfit Recommendation Scheme with Attribute - wise Interpretability,

    D. Sagar, J. Garg, P. Kansal, S. Bhalla, R. R. Shah, and Y. Yu, “PAI - BPR: Personalized Outfit Recommendation Scheme with Attribute - wise Interpretability,” in 2020 IEEE Sixth International Conference on Multimedia Big Data (BigMM), Sep. 2020, pp. 221 –230. doi: 10.1109/BigM...

  56. [65]

    Collaborative Fashion Recommendation: A Functional Tensor Factorization Approach,

    Y. Hu, X. Yi, and L. S. Davis, “Collaborative Fashion Recommendation: A Functional Tensor Factorization Approach,” in Proceedings of the 23rd ACM international conference on Multimedia, in MM ’15. New York, NY, USA: Association for Computing Machinery, Oct. 2015, pp. 129 –138....

  57. [66]

    Modality-Oriented Graph Learning Toward Outfit Compatibility Modeling,

    X. Song, S. -T. Fang, X. Chen, Y. Wei, Z. Zhao, and L. Nie, “Modality-Oriented Graph Learning Toward Outfit Compatibility Modeling,” IEEE Trans. Multimed., vol. 25, pp. 856 –867, 2023, doi: 10.1109/TMM.2021.3134164

  58. [67]

    Personalized Fashion Compatibility Modeling via Metapath -guided Heterogeneous Graph Learning,

    W. Guan, F. Jiao, X. Song, H. Wen, C. -H. Yeh, and X. Chang, “Personalized Fashion Compatibility Modeling via Metapath -guided Heterogeneous Graph Learning,” in Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, in ...

  59. [68]

    GP -BPR: Personalized Compatibility Modeling for Clothing Matching,

    X. Song, X. Han, Y. Li, J. Chen, X. -S. Xu, and L. Nie, “GP -BPR: Personalized Compatibility Modeling for Clothing Matching,” in Proceedings of the 27th ACM International Conference on Multimedia, in MM ’19. New York, NY, USA: Association for Computing Machinery, Oct. 2019, pp...

  60. [69]

    MCCP: multi -modal fashion compatibility and conditional preference model for personalized clothing recommendation,

    Y. Wang, L. Liu, X. Fu, and L. Liu, “MCCP: multi -modal fashion compatibility and conditional preference model for personalized clothing recommendation,” Multimed. Tools Appl., vol. 83, no. 4, pp. 9621–9645, Jan. 2024, doi: 10.1007/s11042-023-15659-5

  61. [70]

    Personalized Outfit Compatibility Prediction Using Outfit Graph Network,

    B. S. Vivek, G. Bhattacharya, J. Gubbi, B. L. V, A. Pal, and P. Balamuralidhar, “Personalized Outfit Compatibility Prediction Using Outfit Graph Network,” in 2023 International Joint Conference on Neural Networks (IJCNN), Gold Coast, Australia: IEEE, Jun. 2023, pp. 1–8. doi: 1...

  62. [71]

    Personalized clothing recommendation fusing the 4 - season color system and users’ biological characteristics,

    X. Su et al., “Personalized clothing recommendation fusing the 4 - season color system and users’ biological characteristics,” Multimed. Tools Appl., vol. 83, no. 5, pp. 12597 –12625, Feb. 2024, doi: 10.1007/s11042-023-16014-4

  63. [72]

    Hierarchical Multimodal Graph Learning for Outfit Compatibility Modelling,

    R. Xu, J. Wang, M. Li, J. Xu, and Y. Li, “Hierarchical Multimodal Graph Learning for Outfit Compatibility Modelling,” IEEE Trans. Emerg. Top. Comput. Intell., vol. 8, no. 6, pp. 4130–4142, Dec. 2024, doi: 10.1109/TETCI.2024.3386774

  64. [73]

    Personalised Outfit Recommendation via History -aware Transformers,

    M. C. Jung, J. Monteil, P. Schulz, and V. Vaskovych, “Personalised Outfit Recommendation via History -aware Transformers,” Sep. 27, 2024, arXiv: arXiv:2407.00289. doi: 10.48550/arXiv.2407.00289

  65. [74]

    PFRS: Personalized Fashion Recommendation System Using EfficientNet,

    A. Shoeb, M. H. Ali, Md. M. Ali, and M. S. Qaseem, “PFRS: Personalized Fashion Recommendation System Using EfficientNet,” in 2024 International Conference on Innovation and Intelligence for Informatics, Computing, and Technologies (3ICT), Nov. 2024, pp. 337–344. doi: 10.1109/3...

  66. [75]

    Leveraging meta-path and co -attention to model consumer preference stability in fashion recommendations,

    Y.-H. Hu, T.-H. Liu, K. Chen, and F. -C. Yeh, “Leveraging meta-path and co -attention to model consumer preference stability in fashion recommendations,” Decis. Support Syst., vol. 194, p. 114455, Jul. 2025, doi: 10.1016/j.dss.2025.114455

  67. [76]

    Discrepancy Learning Guided Hierarchical Fusion Network for Multi-modal Recommendation,

    Y. Dang, Z. Pan, X. Zhang, W. Chen, F. Cai, and H. Chen, “Discrepancy Learning Guided Hierarchical Fusion Network for Multi-modal Recommendation,” Knowl. -Based Syst., vol. 317, p. 113496, May 2025, doi: 10.1016/j.knosys.2025.113496

  68. [77]

    Hashing -Based Efficient Outfit Recommendation,

    W. Guan, X. Song, D. Zhou, and L. Nie, “Hashing -Based Efficient Outfit Recommendation,” in Advanced Multimodal Compatibility Modeling and Recommendation, W. Guan, X. Song, D. Zhou, and L. Nie, Eds., Cham: Springer Nature Switzerland, 2025, pp. 103 –122. doi: 10.1007/978-3-031...

  69. [78]

    StePO -Rec: Towards Personalized Outfit Styling Assistant via Knowledge -Guided Multi -Step Reasoning,

    Y. Bi, Y. Gao, and H. Wang, “StePO -Rec: Towards Personalized Outfit Styling Assistant via Knowledge -Guided Multi -Step Reasoning,” Apr. 14, 2025, arXiv: arXiv:2504.09915. doi: 10.48550/arXiv.2504.09915

  70. [79]

    BPR: Bayesian Personalized Ranking from Implicit Feedback,

    S. Rendle, C. Freudenthaler, Z. Gantner, and L. Schmidt -Thieme, “BPR: Bayesian Personalized Ranking from Implicit Feedback,” May 09, 2012, arXiv: arXiv:1205.2618. doi: 10.48550/arXiv.1205.2618

  71. [80]

    Personalized Fashion Recommendation with Image Attributes and Aesthetics Assessment,

    C. Chen, F. Mo, X. Fan, and H. Yamana, “Personalized Fashion Recommendation with Image Attributes and Aesthetics Assessment,” Jan. 06, 2025, arXiv: arXiv:2501.03085. doi: 10.48550/arXiv.2501.03085

  72. [2024]

    Available: https://openaccess.thecvf.com/content_CVPR_2019/html/Lu_Learnin g_Binary_Code_for_Personalized_Fashion_Recommendation_CVPR _2019_paper.html

    [Online]. Available: https://openaccess.thecvf.com/content_CVPR_2019/html/Lu_Learnin g_Binary_Code_for_Personalized_Fashion_Recommendation_CVPR _2019_paper.html

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.