Pith. sign in

REVIEW 5 major objections 6 minor 50 references

Multi-Modality Transformer for E-Commerce: Inferring User Purchase Intention to Bridge the Query-Product Gap

T0 review · 5 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Pith's one-line read This paper claims that sparse e-commerce queries can be transformed into pseudo-product embeddings, conditioned on crowd-level purchase-intention vectors and granular text and image product features, and that this transformation beats…

desk verdict A promising new combination of purchase-intent centroids and a pseudo-product decoder, but the training objective is under-specified in ways that block reproduction and the evaluation doesn't yet support the full superiority claim. read the letter →

arxiv 2501.14826 v1 pith:KPZVT6JX submitted 2025-01-21 cs.IR cs.AIcs.LG

classification cs.IRcs.AIcs.LG
keywords e-commerceretrievalpurchaseintentionpseudo-productembeddingmulti-modaltransformerreward-basedcompetitivelearningvectorquantizationcontrastiveproductsearch
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

This paper proposes PINCER, a two-stage multi-modal transformer that rewrites a short e-commerce search query into a pseudo-product embedding before matching it against the product catalog. The central claim is that a query alone is too sparse to capture what a user will buy, and that the gap can be closed by conditioning the query on crowd-level purchase-intention vectors learned from query-to-add-to-cart pairs, plus granular text and image product features retrieved at query time. If the claim holds, retrieval systems could surface products that match the underlying intent rather than just the query's surface words, improving early-page recall. The paper reports a 10.81% relative gain in SumR (Recall@10+20+50+100) over RetroMAE on real-world ATC data and consistent gains on synthetic benchmarks.

What carries the argument

The load-bearing object is the purchase-intention vector set $S=\{s_1,\ldots,s_K\}$, a fixed number of learnable centroids in the shared query-product latent space. A reward-based competitive learning rule, adapted from Likas's reinforcement-learning clustering, assigns each query and its add-to-cart product to the nearest intent vector by Euclidean distance, rewards the pair when both choose the same centroid, and moves the centroid toward the pair; the selection probability in Equation (6) acts as a learning-rate modulator. The intention vector then becomes the conditioning signal for the stage-2 transformer decoder, which also cross-attends to query-aligned product text and image features retrieved from a faiss index, and is trained with a preference modeling loss that pulls the pseudo embedding toward the target product while a KL term preserves stage-1 recall.

What would settle it

Replace the hard argmin intent selection with a soft assignment (for instance, a Gumbel-softmax over intent distances) and retrain PINCER under identical settings; if SumR collapses or the intention centroids fail to move from initialization, the non-differentiable competitive update is doing essential work, not the transformer conditioning alone.

Watch

Extended reading notes

Core claim

PINCER's core claim is that sparse user queries can be transformed into pseudo-product embeddings that lie closer to the products users actually add to cart, by injecting two external signals: purchase intentions, defined as shared latent vectors that quantize query-ATC pairs through reward-based competitive learning, and query-aligned granular product text and image features stored in a vector database. Stage 1 trains the query and product encoders with a CLIP-style contrastive loss plus a reward-based competitive learning term that pushes each query and its ATC product to select the same intention vector. Stage 2 freezes those encoders and trains a causal transformer decoder, conditioned on the query, its nearest intention vector, and retrieved product features, to generate a pseudo-product embedding that is optimized by preference modeling against the true target product. The paper argues this two-stage design is what lets the model beat text-only and multi-modal baselines on both real-world and synthetic e-commerce retrieval.

Load-bearing premise

The whole pipeline relies on the hard nearest-intent selection in Equation (4) still providing reliable training gradients, and the paper does not describe how gradients flow through that non-differentiable step; if implemented naively, the intention centroids may never converge and the decoder's conditioning signal would be meaningless.

Editorial extensions

If this is right

  • Queries can be retrieved as pseudo-products: at inference, the decoder produces an embedding compared to catalog product embeddings by cosine similarity, so no query reformulation or pseudo-relevance feedback loop is needed.
  • Early-page recall is where the gain concentrates: PINCER improves Recall@10 and Recall@20 far more than Recall@100 on real-world data, which matters because users mostly buy from the first page.
  • Purchase intention vectors double as cluster centers: the same vectors used for conditioning pre-cluster the product catalog, cutting retrieval latency to 15ms per 1000 queries at a small recall cost.
  • The method works with light encoders: PINCER's gains appear with distilBERT and ResNet-50, not just large pre-trained models, so it can run on modest GPU resources.

Reading between the lines

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

  • If intention-vector conditioning is doing the work, the learned centroids should be interpretable as purchasing contexts (e.g., 'burning-incense gift' vs 'best-sellers incenses'), and the same vectors could explain recommendations or power diversified result sets, which the paper only hints at in its qualitative t-SNE plots.
  • The formulation suggests a testable extension beyond e-commerce: any paired sparse-to-dense retrieval task where the two sides share latent intents could adopt the same two-stage pseudo-target generation.
  • Because intention vectors are shared across users rather than per-user, PINCER sidesteps the cold-start problem of personalized search, but it also cannot adapt to a user whose intent is rare; a per-user refinement of the pseudo-embedding is a natural next step the paper leaves open.
  • The paper does not examine sensitivity to K, the number of intention vectors; a controlled sweep of K on the real-world dataset would reveal whether the performance gain is robust or tuned to a particular quantization granularity.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes PINCER, a two-stage multi-modal transformer for e-commerce product retrieval. In Stage 1, contrastive learning aligns query and product text/image embeddings, while a reward-based competitive learning scheme with vector quantization estimates a set of purchase-intention vectors shared by queries and add-to-cart (ATC) products. In Stage 2, a transformer decoder conditioned on the query embedding, the selected purchase-intention vector, and retrieved granular product features generates a pseudo-product embedding; the decoder is trained with a preference-modeling loss and a KL term intended to preserve Stage 1 recall. The authors report experiments on a proprietary real-world home-decor click-stream dataset and on synthetic datasets derived from Amazon Cross-Market and FashionGen, comparing with RetroMAE, CLIP, and FashionCLIP. They claim consistent recall improvements, including a 10.81% relative SumR gain over RetroMAE on real data, and an ablation study on one synthetic dataset attributes the gains to combining purchase intention and product features.

Significance. If validated, the idea of transforming sparse queries into pseudo-product representations using crowd-derived purchase-intention vectors is a useful contribution to e-commerce retrieval, and the two-stage architecture with vector quantization is a plausible mechanism. The paper makes several strengths explicit: the synthetic datasets are released on OSF, the real-world evaluation uses authentic ATC transaction data, and the comparison includes a text-only strong baseline (RetroMAE). However, the mathematical formulation of both training stages has unresolved issues, the experimental protocol is underpowered (only three baselines, no error bars, no repeated runs, no public code), and the reported statistical significance test appears impossible with the stated sample size. These problems currently prevent the central claims from being accepted as established.

major comments (5)
  1. [§3.2, Eq. (3)–(7)] The reward-based competitive learning loss does not implement the stated negative reward for mismatched query/product intent assignments. For the first term of RCL in Eq. (3), the coefficient of ||xn - skx||^2 is rsk * rpxn,skx. When skx ≠ sky, Eq. (5) gives rsk = -1 and Eq. (7) gives rpxn,skx = -p, so the product is +p > 0. Thus the loss still attracts x to its selected intent vector even for mismatched pairs; there is no repulsive force that separates query and product intent choices. Only the weighting changes. Please correct the sign convention in Eq. (7) or Eq. (3), or state explicitly how the negative reward is realized.
  2. [§3.3, Eqs. (8)–(10)] The Stage 2 objective is not well defined. Eq. (10) defines SD@n,yn as the scalar log-softmax value for the positive pair, i.e., log(exp(@n·yn) / Σ_b exp(@n·yb)). Eq. (8) then takes a KL divergence between two such scalars, which is undefined; KL divergence requires probability distributions. If SD is intended to denote a vector of log-softmax values, the formula in Eq. (10) does not say so. In addition, Eq. (9) has no normalization term, so minimizing PML is equivalent to maximizing an unbounded difference of cosine similarities and is not a proper preference-modeling loss. Please specify the exact distributions and loss terms.
  3. [§3.2 and Algorithm 1, Eq. (4)] The gradient flow through the hard argmin assignment is not specified. In Algorithm 1 line 12, all encoders, projectors, and the intent vectors S are updated using Lstage1, but Lstage1 includes the argmin-selected indices from Eq. (4). The loss in Eq. (3) is differentiable with respect to the selected intent vector, so a detached-assignment / k-means-style update can work, but the paper never states whether assignments are detached, whether a straight-through estimator is used, or whether a soft relaxation is applied. These choices lead to very different training dynamics. The paper also does not report K (the number of intent vectors), the initialization of S, or how empty/dead intent vectors are handled, all of which are necessary for reproducibility.
  4. [§4.5, Wilcoxon test] The claimed Wilcoxon signed-rank test result is impossible with the reported data. The test is run on four synthetic dataset configurations (FashionGen with brightness, FashionGen with mean gradient, Amazon with brightness, Amazon with mean gradient), i.e., n = 4. For a two-sided exact Wilcoxon signed-rank test with n = 4, the minimum attainable p-value is 0.125 (0.0625 one-sided), so p = 0.043114 cannot occur. The statement that PINCER shows statistically significant gains at the 90% confidence level is therefore unsupported. Please provide the exact test statistic, a corrected exact test, or additional independent comparisons.
  5. [§4.1–§4.5] The experimental validation is underpowered for the paper's headline claim. Only three baselines are compared (RetroMAE, CLIP, FashionCLIP); modern e-commerce multi-modal baselines such as CommerceMM, e-CLIP, or EI-CLIP are not included, and the stated reason is that they are not openly available or trainable with the authors' resources. No error bars, multiple seeds, or repeated training runs are reported, so the point estimates in Tables 1 and 2 have no measure of variance. The real-world dataset is proprietary and no code is released, preventing independent replication. Please add at least a small number of repeated runs with variance, stronger baselines where feasible, and a more detailed reproducibility protocol.
minor comments (6)
  1. [Abstract] The abstract says 'This paper propose' but should be 'This paper proposes'.
  2. [Table 2 caption] The caption says 'Comparison of retrieval models on five datasets' but the table reports four synthetic configurations; the real-world results are in Table 1. Please correct the caption.
  3. [Eq. (1)] The outer sum runs to N while the denominator uses batch size B. Please clarify that the sum is over the batch elements and adjust the notation consistently.
  4. [§3, text before Eq. (1)] The notation 'd = 128|256' is unclear; please state the actual embedding dimension used for each dataset and each modality.
  5. [§4.2] The synthetic data generation uses ChatGPT 3.5 with unspecified prompts and an image-brightness/gradient filter for purchase-intention infusion. Please provide the exact prompt template, the number of LLM-generated queries per group, and dataset statistics so that the controlled experiments are reproducible.
  6. [§4.5] The phrase 'at 90% confidence level' should read 'at significance level α = 0.10' or similar, to avoid confusion with confidence intervals.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: PINCER's central retrieval claim is trained and evaluated on held-out query-ATC pairs, so it is not derived from its own fitted parameters by construction.

full rationale

The paper's central claim is that PINCER transforms sparse queries into pseudo-product embeddings using learned purchase-intention vectors and granular product features, and that this improves retrieval over baselines. Nothing in the derivation chain reduces to its inputs by construction. The purchase-intention vectors S are learned from training query-ATC pairs via a reward-based competitive loss, and Stage 2 trains the decoder to map a query, its nearest intent vector, and retrieved product features toward the target product embedding; at inference, the same components produce a pseudo-product embedding that is matched against held-out product embeddings. This is a standard supervised training-and-evaluation loop, not a fitted parameter being renamed as a prediction. The synthetic evaluation is also not circular: the test ATC products are generated by filtering on image brightness/gradient, but the model never sees those filtering labels and must recover the resulting preference pattern from query-product pairs, so the controlled experiment measures learning rather than mere replication of the generative rule. The paper's main technical weakness, the underspecified gradient flow through the hard argmin in Eq. (4) and Algorithm 1 line 12, is a reproducibility and training-dynamics concern, not a circularity: it does not make the evaluation equivalent to the training objective. There are no load-bearing self-citations, no imported uniqueness theorem, and no ansatz smuggled in from prior work by the same authors. Accordingly, no circular step can be exhibited, and the honest finding is no significant circularity.

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The central claim rests on a small set of hand-chosen hyperparameters (K, T, lambda, feature counts, d) and domain assumptions about what ATC pairs and synthetic brightness/gradient rules reveal about purchase intention. The invented latent entity is the purchase-intention centroid, which has no independent evidence outside the paper's own experiments.

free parameters (6)
  • number of purchase intention vectors K (|S|) = not specified
    The count of intent centroids S is a hyperparameter chosen by hand; the paper never states K or a selection criterion, yet the whole pseudo-product generation conditions on the nearest centroid.
  • temperature T in contrastive loss = not specified
    CLIP-style temperature in Eq (1) is a free parameter; no value or schedule is reported.
  • loss weight lambda in Eq (2) = 0.5
    Paper states optimal setting lambda=0.5; this balances contrastive and RCL losses and is fit to performance.
  • number of retrieved product features per query in stage 2 = not specified
    Algorithm 2 retrieves f_jn = argmax_f cos(x_n, f), but multi-feature selection and count are not specified; the decoder input depends on how many text/image feature vectors are concatenated.
  • negative sample y+ selection in preference modeling loss = not specified
    Eq (9) requires nearby negative products y+ from intention clusters; the neighborhood size and sampling strategy are not defined.
  • embedding dimension d = 128
    The shared latent dimension is chosen as 128 for efficiency, a hand-set architectural choice that affects capacity.
assumptions (5)
  • domain assumption A single shared latent space of dimension d can represent queries, products, purchase-intention centroids, and product features such that Euclidean distance and dot-product similarity are meaningful.
    Used throughout Section 3; the entire contrastive and RCL losses assume a shared metric space built from distinct encoders.
  • domain assumption Query to add-to-cart (ATC) pairs in click-stream data are a faithful observable proxy for underlying purchase intention.
    Section 3.2 defines purchase intention as the latent link between a query and its ATC product; if ATC actions are noisy (e.g., cart abandonment, browsing), the learned intent centroids encode noise.
  • domain assumption Pre-trained distilBERT and ResNet-50, with linear projectors, provide sufficient semantic and visual features for e-commerce product matching in a 128-d space.
    Section 4.3 chooses these encoders for efficiency; the framework's performance ceiling depends on their quality.
  • ad hoc to paper Synthetic purchase intentions can be emulated by image brightness and mean gradient filters applied to product images.
    Section 4.2 generates synthetic ATC logs by ranking products and filtering by brightness/gradient; this assumes a simple visual signal captures 'purchase intention', which is a paper-specific construction.
  • ad hoc to paper Hard argmin assignments in Eq (4) provide usable gradients for encoder and centroid updates.
    The reward-based competitive loss updates through non-differentiable nearest-centroid assignments; the paper does not state how gradient flow is handled, which is essential for training stability.
invented entities (1)
  • Purchase intention vector s_k (crowd intent centroids)
    purpose: Latent shared centroid that queries and ATC products are both assigned to; used as conditioning input to the decoder and as cluster centers for fast retrieval.
    Introduced as the paper's core construct; no falsifiable prediction outside the reported retrieval benchmarks, and no external data source verifies that these vectors correspond to human-interpretable intentions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-Modality Transformer for E-Commerce: Inferring User Purchase Intention to Bridge the Query-Product Gap." pith.science (2026). https://pith.science/paper/KPZVT6JX

@misc{pith2026250114826,
  author       = {Pith},
  title        = {Pith review of: Multi-Modality Transformer for E-Commerce: Inferring User Purchase Intention to Bridge the Query-Product Gap},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KPZVT6JX}},
  note         = {Machine review of arXiv:2501.14826}
}
read the original abstract

E-commerce click-stream data and product catalogs offer critical user behavior insights and product knowledge. This paper propose a multi-modal transformer termed as PINCER, that leverages the above data sources to transform initial user queries into pseudo-product representations. By tapping into these external data sources, our model can infer users' potential purchase intent from their limited queries and capture query relevant product features. We demonstrate our model's superior performance over state-of-the-art alternatives on e-commerce online retrieval in both controlled and real-world experiments. Our ablation studies confirm that the proposed transformer architecture and integrated learning strategies enable the mining of key data sources to infer purchase intent, extract product features, and enhance the transformation pipeline from queries to more accurate pseudo-product representations.

Figures

Figures reproduced from arXiv: 2501.14826 by the authors.

Figure 1
Figure 1. Examples to show the importance of purchase intention and product features in users’ [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. PINCER model architecture in various stages of training [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Synthetic Add-To-Cart (ATC) data generation from FashionGen and Amazon datasets using [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: t-SNE distribution of FashionGen(mean gradient) randomly chosen query-product pairs [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Retrieval comparison between RetroMAE, CLIP, and PINCER models for a query from [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Retrieval comparison between RetroMAE, CLIP, and PINCER models for a query from [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Retrieval comparison between full-scale and PINCER clustered products [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: PINCER model architecture 16 [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 44 canonical work pages

  1. [1]

    Learning a hierarchical embedding model for personalized product search

    Qingyao Ai et al. “Learning a hierarchical embedding model for personalized product search”. In: Proceedings of the 40th International ACM SIGIR Conference on Research and Develop- ment in Information Retrieval . 2017, pp. 645–654

  2. [2]

    A general language assistant as a laboratory for alignment

    Amanda Askell et al. “A general language assistant as a laboratory for alignment”. In: arXiv preprint arXiv:2112.00861 (2021)

  3. [3]

    Query expansion techniques for information retrieval: a survey

    Hiteshwar Kumar Azad and Akshay Deepak. “Query expansion techniques for information retrieval: a survey”. In: Information Processing & Management 56.5 (2019), pp. 1698–1735

  4. [4]

    Modeling the impact of short-and long-term behavior on search person- alization

    Paul N Bennett et al. “Modeling the impact of short-and long-term behavior on search person- alization”. In: Proceedings of the 35th international ACM SIGIR conference on Research and development in information retrieval. 2012, pp. 185–194

  5. [5]

    A transformer-based embedding model for personalized product search

    Keping Bi, Qingyao Ai, and W Bruce Croft. “A transformer-based embedding model for personalized product search”. In:Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval . 2020, pp. 1521–1524. 12

  6. [6]

    Cross-Market Product Recommendation

    Hamed Bonab et al. “Cross-Market Product Recommendation”. In: Proceedings of the 30th ACM International Conference on Information & Knowledge Management . ACM, 2021

  7. [7]

    Multi-objective ranking optimization for product search using stochastic label aggregation

    David Carmel et al. “Multi-objective ranking optimization for product search using stochastic label aggregation”. In: Proceedings of The Web Conference 2020. 2020, pp. 373–383

  8. [8]

    Unsupervised learning of visual features by contrasting cluster as- signments

    Mathilde Caron et al. “Unsupervised learning of visual features by contrasting cluster as- signments”. In: Advances in neural information processing systems 33 (2020), pp. 9912– 9924

Show all 50 references
  1. [9]

    Towards a better understanding of query reformulation behavior in web search

    Jia Chen et al. “Towards a better understanding of query reformulation behavior in web search”. In: Proceedings of the web conference 2021 . 2021, pp. 743–755

  2. [10]

    UNITER: UNiversal Image-TExt Representation Learning

    Yen-Chun Chen et al. “UNITER: UNiversal Image-TExt Representation Learning”. In:Com- puter Vision – ECCV 2020 . Ed. by Andrea Vedaldi et al. Cham: Springer International Publish- ing, 2020, pp. 104–120. ISBN : 978-3-030-58577-8

  3. [11]

    Detecting online commercial intention (OCI)

    Honghua Dai et al. “Detecting online commercial intention (OCI)”. In: Proceedings of the 15th international conference on World Wide Web. 2006, pp. 829–837

  4. [12]

    Contrastive Learning for User Sequence Representation in Personalized Product Search

    Shitong Dai et al. “Contrastive Learning for User Sequence Representation in Personalized Product Search”. In: Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 2023, pp. 380–389

  5. [13]

    A large-scale evaluation and analysis of personalized search strategies

    Zhicheng Dou, Ruihua Song, and Ji-Rong Wen. “A large-scale evaluation and analysis of personalized search strategies”. In: Proceedings of the 16th international conference on World Wide Web. 2007, pp. 581–590

  6. [14]

    Semantic models for the first-stage retrieval: A comprehensive review

    Jiafeng Guo et al. “Semantic models for the first-stage retrieval: A comprehensive review”. In: ACM Transactions on Information Systems (TOIS) 40.4 (2022), pp. 1–42

  7. [15]

    Ready to buy or just browsing? Detecting web searcher goals from interaction data

    Qi Guo and Eugene Agichtein. “Ready to buy or just browsing? Detecting web searcher goals from interaction data”. In: Proceedings of the 33rd international ACM SIGIR conference on Research and development in information retrieval . 2010, pp. 130–137

  8. [16]

    Attentive long short-term preference modeling for personalized product search

    Yangyang Guo et al. “Attentive long short-term preference modeling for personalized product search”. In: ACM Transactions on Information Systems (TOIS) 37.2 (2019), pp. 1–27

  9. [17]

    Multi-modal preference modeling for product search

    Yangyang Guo et al. “Multi-modal preference modeling for product search”. In: Proceedings of the 26th ACM international conference on Multimedia . 2018, pp. 1865–1873

  10. [18]

    Momentum contrast for unsupervised visual representation learning

    Kaiming He et al. “Momentum contrast for unsupervised visual representation learning”. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 2020, pp. 9729–9738

  11. [19]

    Analyzing and predicting purchase intent in e-commerce: anony- mous vs. identified customers

    Mariya Hendriksen et al. “Analyzing and predicting purchase intent in e-commerce: anony- mous vs. identified customers”. In: arXiv preprint arXiv:2012.08777 (2020)

  12. [20]

    Extending CLIP for Category-to-image Retrieval in E-commerce

    Mariya Hendriksen et al. “Extending CLIP for Category-to-image Retrieval in E-commerce”. In: European Conference on Information Retrieval. Springer. 2022, pp. 289–303

  13. [21]

    Unsupervised dense information retrieval with contrastive learning

    Gautier Izacard et al. “Unsupervised dense information retrieval with contrastive learning”. In: arXiv preprint arXiv:2112.09118 (2021)

  14. [22]

    Self-supervised product quantization for deep unsu- pervised image retrieval

    Young Kyun Jang and Nam Ik Cho. “Self-supervised product quantization for deep unsu- pervised image retrieval”. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021, pp. 12085–12094

  15. [23]

    Billion-scale similarity search with GPUs

    Jeff Johnson, Matthijs Douze, and Hervé Jégou. “Billion-scale similarity search with GPUs”. In: IEEE Transactions on Big Data 7.3 (2019), pp. 535–547

  16. [24]

    Vilt: Vision-and-language transformer without convolution or region supervision

    Wonjae Kim, Bokyung Son, and Ildoo Kim. “Vilt: Vision-and-language transformer without convolution or region supervision”. In:International Conference on Machine Learning. PMLR. 2021, pp. 5583–5594

  17. [25]

    Automatic identification of user goals in web search

    Uichin Lee, Zhenyu Liu, and Junghoo Cho. “Automatic identification of user goals in web search”. In: Proceedings of the 14th international conference on World Wide Web . 2005, pp. 391–400

  18. [26]

    Align before fuse: Vision and language representation learning with momen- tum distillation

    Junnan Li et al. “Align before fuse: Vision and language representation learning with momen- tum distillation”. In: Advances in neural information processing systems 34 (2021), pp. 9694– 9705

  19. [27]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li et al. “Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation”. In: International Conference on Machine Learning . PMLR. 2022, pp. 12888–12900. 13

  20. [28]

    Embedding-based product retrieval in taobao search

    Sen Li et al. “Embedding-based product retrieval in taobao search”. In:Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining . 2021, pp. 3181–3189

  21. [29]

    A reinforcement learning approach to online clustering

    Aristidis Likas. “A reinforcement learning approach to online clustering”. In: Neural computa- tion 11.8 (1999), pp. 1915–1932

  22. [30]

    Retromae: Pre-training retrieval-oriented transformers via masked auto-encoder

    Zheng Liu and Yingxia Shao. “Retromae: Pre-training retrieval-oriented transformers via masked auto-encoder”. In: arXiv preprint arXiv:2205.12035 (2022)

  23. [31]

    Universal Vision-Language Dense Retrieval: Learning A Unified Rep- resentation Space for Multi-Modal Retrieval

    Zhenghao Liu et al. “Universal Vision-Language Dense Retrieval: Learning A Unified Rep- resentation Space for Multi-Modal Retrieval”. In: The Eleventh International Conference on Learning Representations. 2022

  24. [32]

    Understanding behaviors that lead to purchasing: A case study of pinterest

    Caroline Lo, Dan Frankowski, and Jure Leskovec. “Understanding behaviors that lead to purchasing: A case study of pinterest”. In:Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining . 2016, pp. 531–540

  25. [33]

    Vilbert: Pretraining task-agnostic visiolinguistic representations for vision- and-language tasks

    Jiasen Lu et al. “Vilbert: Pretraining task-agnostic visiolinguistic representations for vision- and-language tasks”. In: Advances in neural information processing systems 32 (2019)

  26. [34]

    EI-CLIP: Entity-Aware Interventional Contrastive Learning for E-Commerce Cross-Modal Retrieval

    Haoyu Ma et al. “EI-CLIP: Entity-Aware Interventional Contrastive Learning for E-Commerce Cross-Modal Retrieval”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022, pp. 18051–18061

  27. [35]

    Semantic product search

    Priyanka Nigam et al. “Semantic product search”. In: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining . 2019, pp. 2876–2885

  28. [36]

    Learning transferable visual models from natural language supervision

    Alec Radford et al. “Learning transferable visual models from natural language supervision”. In: International Conference on Machine Learning . PMLR. 2021, pp. 8748–8763

  29. [37]

    Fashion-gen: The generative fashion dataset and challenge

    Negar Rostamzadeh et al. “Fashion-gen: The generative fashion dataset and challenge”. In: arXiv preprint arXiv:1806.08317 (2018)

  30. [38]

    e-clip: Large-scale vision-language representation learning in e- commerce

    Wonyoung Shin et al. “e-clip: Large-scale vision-language representation learning in e- commerce”. In: Proceedings of the 31st ACM International Conference on Information & Knowledge Management. 2022, pp. 3484–3494

  31. [39]

    Challenges and research opportunities in ecommerce search and recommendations

    Manos Tsagkias et al. “Challenges and research opportunities in ecommerce search and recommendations”. In: ACM SIGIR F orum. V ol. 54. 1. ACM New York, NY , USA. 2021, pp. 1–23

  32. [40]

    Learning latent vector spaces for product search

    Christophe Van Gysel, Maarten de Rijke, and Evangelos Kanoulas. “Learning latent vector spaces for product search”. In: Proceedings of the 25th ACM international on conference on information and knowledge management. 2016, pp. 165–174

  33. [41]

    ONE-PEACE: Exploring One General Representation Model Toward Unlimited Modalities

    Peng Wang et al. “ONE-PEACE: Exploring One General Representation Model Toward Unlimited Modalities”. In: arXiv preprint arXiv:2305.11172 (2023)

  34. [42]

    Pseudo-relevance feedback for multiple representation dense retrieval

    Xiao Wang et al. “Pseudo-relevance feedback for multiple representation dense retrieval”. In: Proceedings of the 2021 ACM SIGIR International Conference on Theory of Information Retrieval. 2021, pp. 297–306

  35. [43]

    The influence of image search intents on user behavior and satisfaction

    Zhijing Wu et al. “The influence of image search intents on user behavior and satisfaction”. In: Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining . 2019, pp. 645–653

  36. [44]

    Improving query representations for dense retrieval with pseudo relevance feedback

    HongChien Yu, Chenyan Xiong, and Jamie Callan. “Improving query representations for dense retrieval with pseudo relevance feedback”. In: arXiv preprint arXiv:2108.13454 (2021)

  37. [45]

    Commercemm: Large-scale commerce multimodal representation learning with omni retrieval

    Licheng Yu et al. “Commercemm: Large-scale commerce multimodal representation learning with omni retrieval”. In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 2022, pp. 4433–4442

  38. [46]

    Product quantization network for fast image retrieval

    Tan Yu et al. “Product quantization network for fast image retrieval”. In:Proceedings of the European Conference on Computer Vision (ECCV). 2018, pp. 186–201

  39. [47]

    Towards personalized and semantic retrieval: An end-to-end solution for e-commerce search via embedding learning

    Han Zhang et al. “Towards personalized and semantic retrieval: An end-to-end solution for e-commerce search via embedding learning”. In: Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval . 2020, pp. 2407– 2416

  40. [48]

    Collaborative quantization for cross-modal similarity search

    Ting Zhang and Jingdong Wang. “Collaborative quantization for cross-modal similarity search”. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition . 2016, pp. 2036–2045

  41. [49]

    Make: Vision-language pre-training based product retrieval in taobao search

    Xiaoyang Zheng et al. “Make: Vision-language pre-training based product retrieval in taobao search”. In: Companion Proceedings of the ACM Web Conference 2023 . 2023, pp. 356–360. 14

  42. [50]

    PSSL: self-supervised learning for personalized search with contrastive sampling

    Yujia Zhou et al. “PSSL: self-supervised learning for personalized search with contrastive sampling”. In: Proceedings of the 30th ACM International Conference on Information & Knowledge Management. 2021, pp. 2749–2758. A Appendix / supplemental material A.1 Algorithm: The quer...

Pith tools

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