Real-time and personalized product recommendations for large e-commerce platforms

Davide Bacciu; Marco Varesi; Matteo Tolloso; Shahab Mokarizadeh

REVIEW 3 major objections 5 minor 26 references

Real-time and personalized product recommendations for large e-commerce platforms

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that distilling a graph recommender into 700 KB per-user models yields 1.5 ms personalized recommendations that beat standard baselines.

desk verdict Sensible HGNN-to-per-user-MLP distillation with an honest ablation, but the headline 1.5 ms / 700 KB efficiency claim is not self-consistent as described. read the letter →

arxiv 2506.21368 v1 pith:KTC6PVJV submitted 2025-06-26 cs.IR cs.AI

Matteo Tolloso , Davide Bacciu , Shahab Mokarizadeh , Marco Varesi This is my paper · ORCID

classification cs.IRcs.AI

keywords RecommendationSystemsGraphNeuralNetworksParsimoniousLearningKnowledgeDistillationContinualPersonalizationE-commerceReal-time

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 claims that a large e-commerce platform can serve personalized product recommendations in about 1.5 milliseconds per user on CPU, with a memory footprint of roughly 700 KB per user, by compressing the knowledge of a heterogeneous graph neural network into tiny per-user models. The compression works by distillation: a global graph model, trained on historical patterns of products being clicked, favorited, added to cart, or bought together, teaches a small image-based model to reproduce its product representations. Each user gets a personal copy of that small model, which is fine-tuned in a few quick steps on that user's recent interactions to reflect current taste. After one week of such adaptation, the paper reports the best F1 score against two standard recommender baselines (LightGCN and PinSage) on a proprietary e-commerce dataset, and better results than the best public solution on a public fashion dataset. If the results hold, the method is a practical way to get real-time personalization under tight latency and memory limits.

What carries the argument

The central object is a three-stage distillation-and-adaptation pipeline. A Heterogeneous Graph Neural Network (HGNN) is first trained with a contrastive loss on a product graph with four edge types—co-click, co-favorite, co-cart, co-purchase—to produce structural item embeddings from ResNet-18 image features. An MLP 'attribute encoder' is then trained via an alignment loss to map those same image embeddings into the HGNN's embedding space without needing graph neighbourhoods, so that at inference time the graph is not required. Finally, each user receives a personal copy of that MLP, updated by a few SGD steps on a triplet loss that pulls positively interacted items toward a weighted centroid of the user's recent interactions and pushes negative examples away; the user's taste is tracked as an exponential moving average of projected embeddings, and the top-K nearest neighbours form the recommendation.

What would settle it

Run the full method comparison on a public dataset that contains click, cart, favorite, and purchase events with timestamps, using the paper's own one-week replay protocol and next-twelve-purchases ground truth; if the method's F1 score after one week of adaptation does not exceed both LightGCN and PinSage, the central superiority claim is refuted. As a second independent check, measure per-user recommendation latency on a production server with many concurrent users and a catalog of millions of items to see whether the 1.5 ms figure persists under load.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a heavy, graph-based recommender can be turned into a set of tiny per-user models without losing accuracy. The pre-trained heterogeneous GNN's embeddings are distilled into an MLP that only needs product-image embeddings, and personalization is achieved by fine-tuning a per-user copy of that MLP with a triplet loss on a weighted centroid of recent interactions. The user's evolving preference is represented as an exponential moving average of projected item embeddings, and recommendations are the nearest neighbours of that vector in the personalized space. The paper reports a recommendation latency around 1.5 ms on CPU, a memory footprint around 700 KB per user, adaptation in 3–150 ms, and, after one week of adaptation, the best F1 score against LightGCN and PinSage on the proprietary dataset while remaining competitive on the public dataset.

Load-bearing premise

The load-bearing premise is that replaying historical user logs as a live stream, with the next twelve actually-purchased articles as ground truth on an undescribed proprietary e-commerce dataset, faithfully measures real-time recommendation quality; if that replay is not representative of live conditions, the reported F1 advantage and latency figures do not transfer.

Editorial extensions

If this is right

Per-user personalization can run on standard CPU hardware in near real time, because the expensive graph computation happens offline and the per-user update costs only a few SGD steps on a shallow MLP.
Cold-start users receive sensible recommendations immediately from the global distilled MLP before any personal adaptation has occurred.
The system's performance peaks after about one week of continuous adaptation and degrades after two or three weeks, implying the personal model has a limited useful lifespan and must be periodically reset or refreshed.
The method transfers to a public dataset, where all adapted variants surpass the best public competition solution, even though that dataset lacks click, cart, and favorite signals.
Because the recommendation step is a nearest-neighbour search, its latency is independent of the model update time, so ongoing personalization does not slow down the live recommendation path.

Reading between the lines

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

Editorial extension: the 700 KB per-user footprint and 1.5 ms CPU latency suggest the personalized model could run on-device or at the network edge, a deployment regime the paper does not test; measuring behaviour under concurrent load and on mobile hardware would be a direct follow-up.
Editorial extension: since the student MLP consumes only product-image embeddings, the pipeline may naturally extend to new catalog items without graph retraining, provided their images are available; the paper does not evaluate this cold-item scenario.
Editorial extension: the observed performance peak at one week of adaptation points to a plasticity-forgetting trade-off; adding a replay memory of past interactions or stronger regularization might extend the useful lifespan, an experiment the paper lists as future work.
Editorial extension: the strongest latency and accuracy claims are anchored to a proprietary dataset that is not publicly described; independent verification on a public stream-replay benchmark with multiple interaction types would clarify how much of the reported advantage is dataset-specific.

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.

Desk Editor's Note

read the letter

Short version: this is a sensible applied-ML integration — distill a heterogeneous GNN into a per-user MLP via knowledge distillation, then fine-tune each user copy online with a triplet loss. The ablation is honest and shows both pieces matter. What the paper does not deliver is a self-consistent account of its headline efficiency numbers.

The core problem is the 1.5 ms / 700 KB claim. Since each user has a personal MLP_u, item embeddings in that user's space can't be shared. Either you compute MLP_u over the full catalog at request time — N forward passes, nowhere near 1.5 ms on CPU for N≈10^5 — or you precompute and store them per user, which is O(N·d) floats. For the public dataset's 105k items and d=64 that's roughly 27 MB per user, not 700 KB. Appendix A.2 says the projected embeddings are 'stored in memory' after adaptation, which makes the 700 KB figure look like it counts only the MLP weights. The only escape is a small candidate set or aggressive dimensionality reduction, and neither is stated. That's a load-bearing gap because the paper's whole selling point is real-time, low-memory personalization.

The empirical section has softer issues. The proprietary 'e-commerce dataset' is never described — no size, no time span, no interaction counts. Three random weeks, no significance tests. The public-dataset gains over the best Kaggle solution are single-digit percentage points, and the baselines are evaluated under a modified protocol, which the authors at least document in Appendix C. None of this is disqualifying on its own; together it means the accuracy claims are believable but not strongly established.

What's genuinely new is the specific combination: HGNN contrastive pre-training, distillation into a lightweight per-user MLP, and continual triplet adaptation. That particular integration isn't in the cited prior work. The ablations clearly support the design, and the writing is candid about limitations, including performance degradation after one or two weeks of personalization.

Who is this for? Someone building a real-time fashion recommender who wants a deployment template with plausible offline numbers. It's not a conceptual advance, but it's a real engineering integration with a clear mechanism.

My recommendation: send it to peer review, but flag the memory/latency inconsistency hard. If the authors can state the candidate-set size or store approximate per-user projections within the claimed budget, the paper is fine. As written, the headline efficiency claim needs a correction.

Referee Report

3 major / 5 minor

Summary. The manuscript proposes a framework for real-time, personalized product recommendations in fashion e-commerce. The architecture combines a heterogeneous GNN trained with relation-wise contrastive losses, a distilled lightweight student MLP that maps CNN image embeddings into the GNN's embedding space, and a per-user personalized MLP that is continually adapted via a triplet loss. The user state is maintained as an EMA of projected product embeddings, and recommendations are produced by K-nearest-neighbor search in the user-specific projected space. The authors evaluate the approach on a proprietary e-commerce dataset and on the public H&M Kaggle dataset, comparing against LightGCN and PinSage, and claim CPU response times of 1.5 ms with 700 KB per-user memory.

Significance. If the reported efficiency and accuracy numbers were reliable, the paper would be a practically valuable contribution: the idea of distilling a heterogeneous GNN into a per-user lightweight MLP with continual triplet-based adaptation is coherent, and the ablation study in Table 4 usefully isolates the contributions of pre-training and personalization. The paper also provides explicit hyperparameter ranges in Appendix B and evaluates on a public benchmark in addition to proprietary data. However, the central claims are not yet established: the primary dataset is never described and no significance tests are reported; the baselines are re-implemented under a modified protocol; and the headline latency/memory numbers are internally inconsistent with the algorithm as written. These are load-bearing issues for the paper's main contribution.

major comments (3)

[Section 2, Eq. (5), Table 1, Appendix A.2] The claimed 1.5 ms CPU latency and 700 KB per-user memory are not self-consistent with the stated recommendation procedure. The KNN search is described as being performed 'in the space projected by MLP_u', and since MLP_u is user-specific, every candidate item embedding MLP_u(h_p^CNN) must be produced separately for each user. Either these embeddings are computed at request time, requiring N forward passes through a multilayer MLP for a catalog of N≈10^5 items (incompatible with 1.5 ms on CPU), or they are precomputed and stored per user, requiring O(N·d) storage that Table 1 does not include. Appendix A.2 says that after adaptation the MLP is used once and 'the embedding produced are stored in memory', but Table 1's 700 KB does not account for this per-user item-embedding store; for N=10^5 and d=64, float32 storage alone is roughly 26 MB per user. The authors should state the candidate-set size used in the experiments and give an exact per-user memory budget, or revise the efficiency claims accordingly.
[Section 3, first paragraph; Table 2] The primary evaluation relies on a proprietary dataset that is never described: no number of users, items, interactions, interaction-type mix, or temporal span is given. Only three random weeks are sampled, with no significance tests across weeks, so the claim that the model 'after one week of adaptation, achieves the best results in F1 score' is not statistically supported; the reported standard deviations are large and the week-to-week differences (e.g., F1 407±56 after 1 week versus 369±41 after 2 weeks) overlap. The paper should provide descriptive statistics for the proprietary dataset, report per-week results, and include paired significance tests (e.g., paired bootstrap or signed-rank tests) for the F1 comparisons.
[Appendix C; Tables 2 and 3] The comparison against LightGCN and PinSage uses a modified protocol in which the baselines are reduced to producing static item embeddings, which are then evaluated with an EMA user vector and KNN search. LightGCN is also trained on item-item graphs rather than its standard user-item bipartite formulation. This is a reasonable attempt at a common evaluation protocol, but it means the results cannot be read as a comparison with the published systems in their native form. The claim that the proposed method outperforms LightGCN and PinSage should be conditioned on this reimplementation, and the authors should justify why the modified protocol is the appropriate common evaluation and report any sensitivity of the baselines to the protocol choices.

minor comments (5)

[Tables 1-4] The table captions are broken by line breaks in the text (e.g., 'T able 1', 'T able 2'), which should be fixed in the final version.
[Section 4] There is a typo in the final paragraph: 'that match the the user's personal style' should read 'that match the user's personal style'.
[Appendix B] The sentence 'All other hyperparameters listed in Tables 5 were optimized using grid search' should refer to 'Table 5' in the singular.
[Section 3, Table 3] The 'Best public solution [11]' baseline is not described; the authors should state what method it corresponds to and how its predictions were obtained.
[Section 2, Eq. (6)] In Eq. (6), the notation B_u is used for a batch of interactions, but the sum is over tuples (h^{CNN}_{p+}, w); the authors should clarify the precise composition of B_u and how the interaction weights w are defined for each event type.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: future-purchase ground truth is independent of the fitted personalization, and the self-citations are background only.

full rationale

The paper's derivation chain is not circular. The HGNN teacher is trained with a contrastive loss (Eq. 2) on historical co-interaction edges, the student MLP is distilled with the alignment loss (Eq. 4), and personalization uses the triplet loss (Eq. 7) on past user interactions. The evaluation metric is explicitly future data: "the ground truth set is composed of the next T articles the user will actually buy," with K=10 and T=12. No fitted parameter is renamed as a prediction, and hyperparameters are selected by grid search on validation data (Appendix B), not on the test window. The triplet loss does use the system's own recommendations as negative examples, but this is implicit negative feedback during training, not a circular derivation of the reported F1 scores. The self-citations [3,14] provide background building blocks such as multi-relational graph processing; they do not carry the central claim, and no uniqueness theorem is imported from prior work. The only serious concern is an internal feasibility tension between the claimed 700 KB per-user memory and the per-user stored projected embeddings implied by Appendix A.2, but this is an engineering consistency issue, not circularity. Therefore the circularity score is 0.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The system depends on several hand-set or grid-searched hyperparameters (interaction weights 4/3/2/1, contrastive loss coefficients, triplet margin, EMA decay, number of SGD steps), all chosen without a reported final configuration. The core assumptions are domain-level: image embeddings are sufficient content signals, the next-T-purchases proxy measures real-time quality, and the co-interaction graph captures useful structure. No invented entities are introduced, so the gravi-ton-style ledger is empty.

free parameters (7)

Contrastive loss weighting gamma1..gamma4 = [1, 0.5, 0.5, 0.1]
Fixed in Appendix B; assigns importance to co-clicked, co-favorite, co-cart, co-purchased relations. Chosen by hand, not derived.
Interaction weights w (purchase/cart/favorite/click) = 4/3/2/1
Defined in Section 2, Eq. (6); hand-set importance of interaction types for the user centroid.
Triplet margin epsilon = not reported (grid: 1, 100, inf)
Grid-searched (Appendix B, Table 5); final value not stated.
EMA decay alpha = not reported (grid: 0.1-1.0)
Grid-searched; final value not stated.
Number of SGD adaptation steps = not reported (grid: 1-80)
Controls plasticity; grid-searched; final value not stated.
HGNN and MLP architecture sizes = not reported (two candidates each)
Grid-searched; final architecture not stated.
Evaluation constants K and T = K=10, T=12
Set 'as common practice' in Section 3; not derived.

assumptions (5)

domain assumption ResNet-18 image embeddings are informative content features for fashion products.
The whole distillation pipeline relies on visual features alone for the student MLP (Section 2 and Figure 2).
domain assumption The next T purchased articles are a valid ground truth for recommendation quality in a session.
Section 3 evaluation protocol; no user study or online A/B test.
domain assumption The historical co-interaction graph (co-clicked, co-favorite, co-cart, co-purchased) captures item-item relations useful for future purchases.
Section 2 graph construction and Eq. (2).
domain assumption Euclidean distance in the adapted embedding space is appropriate for KNN retrieval.
Appendix B, citing [13,16,12] for Euclidean over cosine.
ad hoc to paper Non-interacted recommended products are valid negative examples for the triplet loss.
Section 2, negative sampling for Eq. (7); this is a self-referential training signal from the system's own suggestions.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Real-time and personalized product recommendations for large e-commerce platforms." pith.science (2026). https://pith.science/paper/KTC6PVJV

@misc{pith2026250621368,
  author       = {Pith},
  title        = {Pith review of: Real-time and personalized product recommendations for large e-commerce platforms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KTC6PVJV}},
  note         = {Machine review of arXiv:2506.21368}
}

read the original abstract

We present a methodology to provide real-time and personalized product recommendations for large e-commerce platforms, specifically focusing on fashion retail. Our approach aims to achieve accurate and scalable recommendations with minimal response times, ensuring user satisfaction, leveraging Graph Neural Networks and parsimonious learning methodologies. Extensive experimentation with datasets from one of the largest e-commerce platforms demonstrates the effectiveness of our approach in forecasting purchase sequences and handling multi-interaction scenarios, achieving efficient personalized recommendations under real-world constraints.

Figures

Figures reproduced from arXiv: 2506.21368 by the authors.

**Figure 2.** Overall architecture of our proposed recommendation framework, consisting of three core components: (i) the structural encoder (HGNN), which uses a Heterogeneous Graph Neural Network to process the historical interaction graph and capture intricate relational information between products; (ii) the attribute encoder (Student MLP), a lightweight Multi-Layer Perceptron trained via knowledge distillation to approximat… view at source ↗

**Figure 3.** Distillation process. On the upper part the GNN takes in input the CNN embeddings and the graph structure. On the bottom part the MLP takes in input only the CNN embeddings. The alignment loss pushes the two embeddings spaces to be similar. Note that the embeddings’ dimensionality is also compressed by the two networks. The HGNN serves as a teacher model to transfer knowledge to the lightweight student model, i.e. t… view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Continual personalization process within the recommendation framework. When a user interacts with products (green points), the user’s personal embedding space is adjusted so that these positively interacted items move closer to a calculated user preference centroid. Co…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 21 canonical work pages

[1]

Neural Networks129, 203–221 (2020)

Bacciu, D., Errica, F., Micheli, A., Podda, M.: A gentle introduction to deep learn- ing for graphs. Neural Networks129, 203–221 (2020)

work page 2020
[2]

Personalized Audiobook Recommendations at Spotify Through Graph Neural Networks

De Nadai, M., Fabbri, F., Gigioli, P., Wang, A., Li, A., Silvestri, F., Kim, L., Lin, S., Radosavljevic, V., Ghael, S., et al.: Personalized audiobook recommendations at spotify through graph neural networks. arXiv preprint arXiv:2403.05185 (2024)

work page Pith review arXiv 2024
[3]

Neurocomputing 504, 42–55 (2022) 12 M

Dukic, H., Mokarizadeh, S., Deligiorgis, G., Sepe, P., Bacciu, D., Trincavelli, M.: Inductive-transductive learning for very sparse fashion graphs. Neurocomputing 504, 42–55 (2022) 12 M. Tolloso et al

work page 2022
[4]

ACM Transactions on Recommender Systems 1(1), 1–51 (2023)

Gao, C., Zheng, Y., Li, N., Li, Y., Qin, Y., Piao, J., Quan, Y., Chang, J., Jin, D., He, X., et al.: A survey of graph neural networks for recommender systems: Challenges, methods, and directions. ACM Transactions on Recommender Systems 1(1), 1–51 (2023)

work page 2023
[5]

Inter- national Journal of Computer Vision129(6), 1789–1819 (2021)

Gou, J., Yu, B., Maybank, S.J., Tao, D.: Knowledge distillation: A survey. Inter- national Journal of Computer Vision129(6), 1789–1819 (2021)

work page 2021
[6]

Advances in neural information processing systems30 (2017)

Hamilton, W., Ying, Z., Leskovec, J.: Inductive representation learning on large graphs. Advances in neural information processing systems30 (2017)

2017
[7]

Inductive Link Prediction for Nodes Having Only Attribute Information

Hao, Y., Cao, X., Fang, Y., Xie, X., Wang, S.: Inductive link prediction for nodes having only attribute information. arXiv preprint arXiv:2007.08053 (2020)

work page Pith review arXiv 2020
[8]

He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

2016

Show all 26 references

[9]

In: Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval

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

2020
[10]

H&M: Competition on personalization,https://www.kaggle.com/competitions/ h-and-m-personalized-fashion-recommendations/data
[11]

JacobCP: https://github.com/JacobCP/kaggle-handm-helpers
[12]

The Programming Historian (2020)

Ladd, J.R.: Understanding and using common similarity measures for text analysis. The Programming Historian (2020)

2020
[13]

Indian Economic Review58(2), 427–431 (2023)

Mukherjee, S., Sonal, R.: A reconciliation between cosine similarity and euclidean distance in individual decision-making problems. Indian Economic Review58(2), 427–431 (2023)

2023
[14]

In: 2024 International Joint Confer- ence on Neural Networks (IJCNN)

Sattar, A., Deligiorgis, G., Trincavelli, M., Bacciu, D.: Multi-relational graph neu- ral network for out-of-domain link prediction. In: 2024 International Joint Confer- ence on Neural Networks (IJCNN). pp. 1–8 (2024).https://doi.org/10.1109/ IJCNN60899.2024.10650198

2024
[15]

ACM Computing Surveys 55(13s), 1–40 (2023)

Song, Y., Wang, T., Cai, P., Mondal, S.K., Sahoo, J.P.: A comprehensive survey of few-shot learning: Evolution, applications, challenges, and opportunities. ACM Computing Surveys 55(13s), 1–40 (2023)

2023
[16]

arXiv preprint arXiv:2407.08623 (2024)

Tessari, F., Hogan, N.: Surpassing cosine similarity for multidimensional comparisons: Dimension insensitive euclidean metric (diem). arXiv preprint arXiv:2407.08623 (2024)

2024 arXiv
[17]

Vettoruzzo, A., Bouguelia, M.R., Vanschoren, J., Rognvaldsson, T., Santosh, K.: Advancesandchallengesinmeta-learning:Atechnicalreview.IEEETPAMI(2024)

2024
[18]

IEEE TPAMI (2024)

Wang, L., Zhang, X., Su, H., Zhu, J.: A comprehensive survey of continual learning: theory, method and application. IEEE TPAMI (2024)

2024
[19]

In: Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining

Ying, R., He, R., Chen, K., Eksombatchai, P., Hamilton, W.L., Leskovec, J.: Graph convolutional neural networks for web-scale recommender systems. In: Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. pp. 974–983 (2018)

2018
[20]

In: Proceedings of the 25th ACM SIGKDD international confer- ence on knowledge discovery & data mining

Zhang, C., Song, D., Huang, C., Swami, A., Chawla, N.V.: Heterogeneous graph neural network. In: Proceedings of the 25th ACM SIGKDD international confer- ence on knowledge discovery & data mining. pp. 793–803 (2019) Real-time and personalized product recommendations 13 A Addit...

2019
[21]

Thesetofnodesispartitionedinbatchessuchthateachbatchhas batch_size nodes
[22]

Starting from each node in the batch, iteratively sample some edges giving more probability to edges with higher weighs
[23]

Add the nodes to the other end of the selected edges (and the edges them- selves) to the graph batch. The number of iterations and the number of edges to add at each iteration is described by thenum_neighbor list, for example the list[3, 2]indicates to sample three edges for e...

2020
[24]

The user’s evolving interest within a session is represented by an Exponential Moving Average (EMA) of the learned item embeddings corresponding to the items interacted with so far in that session
[25]

Recommendations are generated by performing aK-Nearest Neighbor (K- NN) search against this session-specific EMA user profile vector within the global space of learned item embeddings
[26]

The same session-based Precision, Recall, and F1-score metrics (comparing against the nextT items, as detailed in Section 3) are then applied

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Neural Networks129, 203–221 (2020)

Bacciu, D., Errica, F., Micheli, A., Podda, M.: A gentle introduction to deep learn- ing for graphs. Neural Networks129, 203–221 (2020)

work page 2020

[2] [2]

Personalized Audiobook Recommendations at Spotify Through Graph Neural Networks

De Nadai, M., Fabbri, F., Gigioli, P., Wang, A., Li, A., Silvestri, F., Kim, L., Lin, S., Radosavljevic, V., Ghael, S., et al.: Personalized audiobook recommendations at spotify through graph neural networks. arXiv preprint arXiv:2403.05185 (2024)

work page Pith review arXiv 2024

[3] [3]

Neurocomputing 504, 42–55 (2022) 12 M

Dukic, H., Mokarizadeh, S., Deligiorgis, G., Sepe, P., Bacciu, D., Trincavelli, M.: Inductive-transductive learning for very sparse fashion graphs. Neurocomputing 504, 42–55 (2022) 12 M. Tolloso et al

work page 2022

[4] [4]

ACM Transactions on Recommender Systems 1(1), 1–51 (2023)

Gao, C., Zheng, Y., Li, N., Li, Y., Qin, Y., Piao, J., Quan, Y., Chang, J., Jin, D., He, X., et al.: A survey of graph neural networks for recommender systems: Challenges, methods, and directions. ACM Transactions on Recommender Systems 1(1), 1–51 (2023)

work page 2023

[5] [5]

Inter- national Journal of Computer Vision129(6), 1789–1819 (2021)

Gou, J., Yu, B., Maybank, S.J., Tao, D.: Knowledge distillation: A survey. Inter- national Journal of Computer Vision129(6), 1789–1819 (2021)

work page 2021

[6] [6]

Advances in neural information processing systems30 (2017)

Hamilton, W., Ying, Z., Leskovec, J.: Inductive representation learning on large graphs. Advances in neural information processing systems30 (2017)

2017

[7] [7]

Inductive Link Prediction for Nodes Having Only Attribute Information

Hao, Y., Cao, X., Fang, Y., Xie, X., Wang, S.: Inductive link prediction for nodes having only attribute information. arXiv preprint arXiv:2007.08053 (2020)

work page Pith review arXiv 2020

[8] [8]

He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

2016