Pith. sign in

REVIEW 4 major objections 7 minor 55 references

Towards Unifying Feature Interaction Models for Click-Through Rate Prediction

T0 review · 4 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read The paper proposes IPA, a three-component framework that unifies most explicit feature-interaction models for click-through rate prediction, and derives a new model, PFL, that is competitive with state-of-the-art models and has been…

desk verdict Useful framework and honest results, but the component-wise findings are thinner than the prose, and Finding 4 is confounded. read the letter →

arxiv 2411.12441 v1 pith:MNH3S75A submitted 2024-11-19 cs.IR

classification cs.IR
keywords CTRpredictionfeatureinteractionfactorizationmachinesIPAframeworkembeddingcollapsedimensionalrecommendersystemsproductiondeployment
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

Most explicit feature-interaction models for click-through rate prediction, from classic factorization machines to modern cross networks, share a hidden modular structure. The paper proposes IPA, a framework that decomposes any such model into three choices: the Interaction Function (how two embedding vectors are combined), the Layer Pooling (how higher-order interaction layers are built), and the Layer Aggregator (how layers are combined before the classifier). It argues that most existing models, including FM, FwFM, FvFM, FmFM, xDeepFM, and DCN V2, are specific settings of these three components. Under controlled comparisons, richer interaction functions improve accuracy and reduce embedding collapse, field-wise pooling outperforms global pooling, and layer-wise aggregation lets a model learn the true order of the data. Following these findings, the paper derives PFL, a Projected product with Field pooling and Layer aggregation, which matches state-of-the-art models on public benchmarks and produced significant online GMV lifts in an advertising platform.

What carries the argument

The load-bearing object is the interaction matrix $\boldsymbol{W}$ with its four canonical forms: identity (Naive), scaled identity (Weighted), diagonal (Diagonal), and full (Projected). Along with the Field/Global pooling recursion and the Direct/Layer/Term/Element aggregators, this matrix defines the entire vocabulary of the IPA framework. The component analysis carries the argument: fixing two components isolates the effect of the third, so the reported rankings are causal claims inside the framework. The layer-wise scalar $\alpha_l$ in the Layer Aggregator is the mechanism that lets PFL learn data order by down-weighting redundant high-order layers.

What would settle it

A controlled sweep across all component combinations on a public CTR dataset would settle the ranking claims: if any pair of models that differ only in the Interaction Function reverses the ordering Projected, Diagonal, Weighted, Naive in AUC, or if Field pooling is beaten by Global pooling under a fixed aggregator, the framework's component-wise conclusions fail. A single published model that cannot be expressed as a component triplet without ad hoc changes would falsify the unification claim.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the interaction module of a CTR model can be written as a triple (Interaction Function, Layer Pooling, Layer Aggregator). The Interaction Function is parameterized by a matrix $\boldsymbol{W}\in\mathbb{R}^{K\times K}$ through $f(\boldsymbol{t}_i,\boldsymbol{t}_j,\boldsymbol{W}) = (\boldsymbol{t}_i^\top\boldsymbol{W})\odot \boldsymbol{t}_j^\top$, with four realized forms: Naive (identity), Weighted (scaled identity), Diagonal, and Projected (full matrix). Layer Pooling builds order-$l$ terms either per field or globally, and the Layer Aggregator combines layers by Direct, Layer-wise, Term-wise, or Element-wise weighting. The paper reports that with the other components fixed, more complex interaction matrices consistently improve AUC and LogLoss and also make learned embeddings less collapsed, that Field pooling beats Global pooling, and that layer-wise aggregation lets a model identify the data's interaction order and ignore redundant layers. The derived PFL model (Projected + Field + Layer) reaches AUC 0.8138 on Criteo and 0.7916 on Avazu, is competitive with DCN V2, and showed 0.9%-3.7% GMV lifts in an online A/B test before production deployment.

Load-bearing premise

The framework assumes that every relevant CTR interaction model can be faithfully described by choosing one Interaction Function, one Layer Pooling, and one Layer Aggregator, without re-defining those components for each model.

Editorial extensions

If this is right

  • FM, FwFM, FvFM, FmFM, xDeepFM, and DCN V2 become specific component triplets, so the framework turns model comparisons into controlled experiments.
  • Richer interaction matrices (Projected over Diagonal over Weighted over Naive) yield better AUC and LogLoss under fixed pooling and aggregator, and also produce less collapsed embeddings.
  • Field-wise pooling outperforms global pooling across the tested configurations, pointing to redundancy in global pooling as a performance bottleneck.
  • Layer-wise aggregation gives the model the ability to learn the interaction order of the data, keeping performance stable when extra layers are added.
  • PFL, the derived model, is competitive with DCN V2 on Criteo and Avazu, and its online A/B test showed 0.9%-3.7% GMV lifts, leading to production deployment.

Reading between the lines

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

  • Beyond the paper: if the IPA vocabulary is extended to attention-based interaction modules, treating attention as a structured Interaction Function, the same controlled-comparison methodology could test whether attention mechanisms subsume or outperform the four matrix forms.
  • Beyond the paper: the reported Interaction-Function ranking is a target for replication on additional datasets and embedding sizes; a reversal would show the ranking is dataset-dependent rather than a universal design law.
  • Beyond the paper: making the layer-wise weights $\alpha_l$ input-dependent is a natural extension of the order-learning finding, since the true interaction order of a query may vary from instance to instance.
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

4 major / 7 minor

Summary. The paper proposes IPA, a three-component framework (Interaction Function, Layer Pooling, Layer Aggregator) intended to unify explicit feature-interaction models for CTR prediction. It maps existing models such as FM, FwFM, FvFM, FmFM, xDeepFM, FiBiNet, and DCN V2 to component choices, conducts component-wise comparisons on Criteo and Avazu, analyzes embedding collapse, and derives a new model PFL (Projected Product, Field Pooling, Layer Aggregator). The paper reports that PFL is competitive with DCN V2 on public benchmarks, summarizes a synthetic-data experiment on learning interaction order, and reports a production A/B test with GMV lifts in Tencent's advertising platform.

Significance. If the component-level findings hold, the IPA framework would provide a useful organizing viewpoint for a crowded model family, and the derived PFL model would be a practical contribution backed by production deployment. The paper's strengths include the use of public datasets with repeated runs, a concrete new model, and an online A/B test as independent evidence. The main weaknesses are that the causal component findings are not all rigorously isolated: the synthetic experiment for the Layer Aggregator is confounded with residual connections, Finding 1 lacks significance testing and conflates structural complexity with parameter count, and the claimed coverage of the framework is asserted rather than delimited. These issues matter because the paper's design guidance and the motivation for PFL rest on the component-wise findings.

major comments (4)
  1. [§3.5, §2.5, and §A.2.4 (Eq. 12)] The synthetic experiment in Figures 5(c) and 5(d) does not isolate the effect of the Layer Aggregator. As Section 2.5 states, PFL differs from DCN V2 in two ways: it removes the residual connection and it adds layer-wise weights. Equation (12) shows that CrossNet's residual term t_{l-1,n} carries lower-order information into every higher layer, which changes information flow and optimization independently of alpha_l. The text then attributes the performance gap to 'the introduction of alpha_l' and draws Finding 4, but the comparison of CIN, CrossNet, and PFL is a comparison in which two components change simultaneously. The paper already includes PFD (Projected, Field, Direct) in Table 3, so a 2x2 ablation on the synthetic data with factors {Direct, Layer} x {with residual, without residual} is needed to support the claim that the layer-wise aggregator filters redundant layers. This is load-bearing because Finding 4 is one of the three component conclusions used to select PFL's design.
  2. [§3.2, Finding 1, Table 3 and Figure 2(a)] Finding 1 is stated as a universal monotonic trend ('the more complicated the projection matrix ... the better the results'), but the evidence is a small set of comparisons with three repeats and no significance tests. More importantly, the four Interaction Functions differ not only in structural form but also in parameter count: Projected Product has O(M K^2) parameters, Diagonal O(M K), Weighted O(M), and Naive none, so the observed ordering could be at least partly a capacity effect. The paper should report paired significance statistics (e.g., t-tests or bootstrap intervals over the repeated runs) and, if the claim is that the projection structure itself is beneficial, include a matched-capacity control or an explicit discussion of why parameter count can be treated as part of the 'complexity' being tested.
  3. [§2.1, Table 1, and §2.4] The unification claim in the abstract and Section 2.4 says that 'most existing models can be categorized within our framework,' but the framework's Interaction Function is restricted to bilinear forms (t_i^T W) ⊙ t_j^T with one of four matrix types. This excludes interaction functions such as attention-weighted interactions (e.g., AFM) or multiplicative blocks (e.g., MaskNet), and the paper provides no formal criterion for what 'most' means or how the four matrix types were shown to be exhaustive for the models claimed to be covered. Because the component-wise findings are controlled experiments only within the expressible family, the coverage claim should be made precise or softened; as written, the 'unification' is a useful taxonomy but not a demonstrated completeness result.
  4. [§3.3, Figures 3 and 4] The dimensional-collapse analysis in Section 3.3 compares FM, FwFM, and FmFM and attributes the differences solely to the Interaction Function. While the Layer Pooling and Layer Aggregator are indeed fixed, the models also differ in the number of trainable parameters by design, so the singular-value differences may reflect capacity or optimization effects rather than a property of the projection structure per se. The figures are qualitative and no variance or error bars are reported. Since Finding 2 is used as a mechanism for Finding 1, the authors should either add a parameter-matched or regularization-controlled comparison, or explicitly frame the collapse result as descriptive rather than causal.
minor comments (7)
  1. [Table 2] The output row for Direct Aggregator, r = Σ_{l=1}^L ∥_{n=1}^M {t_{l,n}}, is ambiguous: it appears to sum concatenated per-layer vectors rather than concatenating the layers, which contradicts the definition 'Directly link each layer' and would also change the output dimensionality. Please correct the notation.
  2. [Eq. (4)] The indices in Equation (4) are inconsistent: alpha_{m,n} weights f(t_n, t_{l-1,m}) but the summation variables m and n are reused in a way that makes the intended pairing unclear. Aligning the indices with Equations (5) and (6) would improve readability.
  3. [§2.2.2] The sentence 'xDeepFM employs AGT to construct layers' uses the undefined abbreviation AGT; it should be CIN or the full name of the compressed interaction network should be given.
  4. [Table 3] The parenthetical values (e.g., '2e-4') appear to be standard deviations over three runs, but the caption does not state this. Please state the exact statistic and the number of runs in the caption or in Section 3.1.4.
  5. [§3.4 and Figure 2(b)] The claim that Field Pooling 'constantly outperforms' Global Pooling is not fully supported by the rows shown in Table 3, which contains no WGL, DGL, or PGL variants with the same Layer Aggregator. Please add the missing rows or clearly identify which curves in Figure 2(b) correspond to those variants and report the H values used for Global Pooling in the figure or legend.
  6. [References] References [13] and [14] are duplicate entries for the same paper ('On the Embedding Collapse when Scaling up Recommendation Models'). Only one citation should be kept, with the other reference renumbered or replaced by the intended distinct work.
  7. [§3.7] The online A/B section reports four GMV lift percentages but lists a different number of scenario names, and it does not provide confidence intervals or p-values for the t-tests. Clarifying the scenario-to-lift mapping and reporting the uncertainty would make the production evidence easier to evaluate.

Circularity Check

1 steps flagged · score 5.0 of 10

PFL's component choices are selected on the same Criteo/Avazu test sets used to report its performance, so the public-benchmark result is partly a selection artifact; the IPA unification itself is not circular.

  1. fitted input called prediction [Section 3.6 (RQ5), building on Sections 3.2, 3.4, 3.5 and Table 3]
    "Based on the above evaluation, the Projected Product and Field Pooling outperforms other choices while Layer Agg. outperforms other aggregators in the condition of previous two components, we derive a new model by choosing these most powerful choice within each component, and name the new model as PFL. ... On both datasets, PFL achieves competitive performance with DCN V2."

    The 'above evaluation' is the same Criteo-x1/Avazu comparison shown in Table 3 and Figure 2, in which PFL's own results already appear. The three PFL components are selected because they were the best or tied-best performers on these exact test sets, and then PFL's AUC/Logloss on the same test sets is reported as evidence that the derived model is competitive. The reported performance is therefore partly a model-selection result: the architecture was chosen to maximize performance on the benchmark used for the evaluation, so the benchmark numbers do not independently test the design rule. The online A/B test provides separate evidence, but the public-table result remains partly circular.

full rationale

The central IPA unification itself is not circular: the mappings of FM, FwFM, FvFM, FmFM, xDeepFM, and DCN V2 to component triples (Appendix A.2) are algebraic rewritings of those models' published update equations, so the taxonomy is checked against external model definitions rather than against the paper's own conclusions. Findings 1-3 are ordinary empirical comparisons, not tautologies; they could have failed. The main circularity is the derivation/evaluation loop for PFL: Section 3.6 chooses Projected Product, Field Pooling, and Layer Agg. as 'the most powerful' components on the basis of the same Criteo/Avazu experiments, and Table 3 then uses results on those same test sets as the evidence that PFL is competitive. This makes the headline public-benchmark result partly a selection artifact. The online A/B test is independent and mitigates the concern, though it is not documented with experimental detail. Finding 4's synthetic experiment is additionally confounded because PFL differs from DCN V2 in two ways per the paper's own Section 2.5: it removes residual connections and changes the aggregator, so the performance gap cannot be attributed to alpha_l alone; this is a validity flaw rather than a by-construction circularity, and it weakens the 'Layer Agg. outperforms' premise used in the selection step. Self-citations such as [13,14,27,34] provide the embedding-collapse metric and production architecture, but the paper re-implements the metric and does not use those citations to force the central derivation, so they are not load-bearing circularity.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The paper introduces a taxonomy and a model variant rather than new physical or mathematical entities. The free parameters are experimental hyperparameters, and the axioms are the domain assumptions underlying the classification and the empirical findings. The central claim rests on the completeness of the proposed vocabulary, which is asserted rather than proven.

free parameters (4)
  • Number of interaction layers L = L=4 for Criteo, L=5 for Avazu
    Set based on best validation performance; affects all high-order model comparisons in Table 3.
  • Global pooling width H = 10 for derived global-pooling models, tuned over 5, 10, 20
    Hyperparameter for global pooling layer terms; results in Fig. 2(b).
  • Embedding size K = 16
    Standard hyperparameter, affects capacity of all models and the interaction matrices.
  • Dropout and learning rate = dropout 0.2, Adam lr 0.001
    Training hyperparameters, standard but affect all results.
assumptions (6)
  • domain assumption Every considered CTR interaction module can be represented as a combination of one of the four Interaction Functions (Naive, Weighted, Diagonal, Projected), one of two Layer Pooling strategies (Field or Global), and one Layer Aggregator.
    Central premise of the unification; stated in Sec. 2, not proved exhaustively.
  • domain assumption The Interaction Function of models can be written as f(t_i, t_j, W) = (t_i^T W) elementwise-multiplied by t_j^T for a matrix W.
    Sec. 2.1 Eq. 2; used to classify all models.
  • domain assumption A more complex projection matrix (full larger than diagonal larger than scaled identity larger than identity) is the cause of both better accuracy and less embedding collapse.
    Findings 1 and 2 infer causation from correlation; confounding factors such as parameter count or training dynamics are not controlled.
  • domain assumption Synthetic data generated by Eq. 8 with monomial cross-terms captures the true interaction order of real CTR data.
    Sec. 3.1.2; used to conclude that PFL can learn data order.
  • domain assumption Layer weight alpha_l in PFL can be analyzed separately from W_l to reflect layer importance.
    Sec. 3.5 acknowledges alpha_l may be absorbed by W_l, yet uses alpha_l times ||W_{l-1}||_F to infer layer contribution.
  • domain assumption The reported online A/B t-tests and GMV lifts are correct.
    Sec. 3.7; no methodology or data is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Unifying Feature Interaction Models for Click-Through Rate Prediction." pith.science (2026). https://pith.science/paper/MNH3S75A

@misc{pith2026241112441,
  author       = {Pith},
  title        = {Pith review of: Towards Unifying Feature Interaction Models for Click-Through Rate Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MNH3S75A}},
  note         = {Machine review of arXiv:2411.12441}
}
read the original abstract

Modeling feature interactions plays a crucial role in accurately predicting click-through rates (CTR) in advertising systems. To capture the intricate patterns of interaction, many existing models employ matrix-factorization techniques to represent features as lower-dimensional embedding vectors, enabling the modeling of interactions as products between these embeddings. In this paper, we propose a general framework called IPA to systematically unify these models. Our framework comprises three key components: the Interaction Function, which facilitates feature interaction; the Layer Pooling, which constructs higher-level interaction layers; and the Layer Aggregator, which combines the outputs of all layers to serve as input for the subsequent classifier. We demonstrate that most existing models can be categorized within our framework by making specific choices for these three components. Through extensive experiments and a dimensional collapse analysis, we evaluate the performance of these choices. Furthermore, by leveraging the most powerful components within our framework, we introduce a novel model that achieves competitive results compared to state-of-the-art CTR models. PFL gets significant GMV lift during online A/B test in Tencent's advertising platform and has been deployed as the production model in several primary scenarios.

Figures

Figures reproduced from arXiv: 2411.12441 by the authors.

Figure 1
Figure 1. Illustration of the IPA framework and the common choices of its three components. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Performance of various choices within each com [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Comparison of embedding collapse for 2-order interaction models on the Criteo-x1 dataset. Each of singular value sum [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Field-wise singular value spectrum for 2-order in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Trends of 𝛼𝑙 , 𝛼𝑙 ∗ ∥𝑾𝑙−1 ∥F and model performances in the training process. Our model learns low 𝛼𝑙 and ∥𝑾𝑙−1 ∥F for extra layers (5-10), obtaining high-level and robust performance even when over-estimating data order. Finding 3. Under the same setting of feature Int…
Figure 6
Figure 6. Figure 6: Dimensional Collapse of IPNN v.s. PFL in our sys [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 27 canonical work pages

  1. [1]

    Avazu Dataset

    2014. Avazu Dataset. https://www.kaggle.com/competitions/avazu-ctr- prediction/data

  2. [2]

    Criteo Dataset

    2014. Criteo Dataset. https://www.kaggle.com/c/criteo-display-ad-challenge/ data

  3. [3]

    Vito Walter Anelli, Alejandro Bellogín, Tommaso Di Noia, and Claudio Pomo

  4. [4]

    Alex Beutel, Paul Covington, Sagar Jain, Can Xu, Jia Li, Vince Gatto, and Ed H Chi. 2018. Latent cross: Making use of context in recurrent recommender systems. In WSDM. 46–54

  5. [5]

    Mathieu Blondel, Akinori Fujino, Naonori Ueda, and Masakazu Ishihata. 2016. Higher-order factorization machines. NeurIPS 29 (2016)

  6. [6]

    2024.IAB/PwC Internet Advertising Revenue Report

    Interactive Advertising Bureau. 2024.IAB/PwC Internet Advertising Revenue Report

  7. [7]

    Yin-Wen Chang, Cho-Jui Hsieh, Kai-Wei Chang, Michael Ringgaard, and Chih-Jen Lin. 2010. Training and testing low-degree polynomial data mappings via linear SVM. Journal of machine learning research 11, Apr (2010), 1471–1490

  8. [8]

    Olivier Chapelle, Eren Manavoglu, and Romer Rosales. 2015. Simple and scalable response prediction for display advertising. ACM transactions on intelligent systems and technology (TIST) 5, 4 (2015), 61

Show all 55 references
  1. [9]

    Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al

  2. [10]

    Yuan Cheng and Yanbo Xue. 2021. Looking at CTR Prediction Again: Is Attention All You Need?. In SIGIR. 1279–1287

  3. [11]

    Ningya Feng, Junwei Pan, Jialong Wu, Baixu Chen, Ximei Wang, Qian Li, Xian Hu, Jie Jiang, and Mingsheng Long. 2024. Long-Sequence Recommendation Models Need Decoupled Embeddings. arXiv preprint arXiv:2410.02604 (2024)

  4. [12]

    Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. 2017. DeepFM: a factorization-machine based neural network for CTR prediction.arXiv preprint arXiv:1703.04247 (2017)

  5. [14]

    Xingzhuo Guo, Junwei Pan, Ximei Wang, Baixu Chen, Jie Jiang, and Mingsheng Long. 2023. On the Embedding Collapse when Scaling up Recommendation Models. arXiv preprint arXiv:2310.04400 (2023)

  6. [15]

    Xiangnan He and Tat-Seng Chua. 2017. Neural factorization machines for sparse predictive analytics. In SIGIR. 355–364

  7. [16]

    Tongwen Huang, Zhiqi Zhang, and Junlin Zhang. 2019. FiBiNET: combining fea- ture importance and bilinear feature interaction for click-through rate prediction. In Recsys. 169–177

  8. [17]

    Li Jing, Pascal Vincent, Yann LeCun, and Yuandong Tian. 2022. Understanding Dimensional Collapse in Contrastive Self-supervised Learning. arXiv preprint arXiv:2110.09348 (2022)

  9. [18]

    Yuchin Juan, Yong Zhuang, Wei-Sheng Chin, and Chih-Jen Lin. 2016. Field-aware factorization machines for CTR prediction. In RecSys. 43–50

  10. [19]

    Yehuda Koren, Robert Bell, and Chris Volinsky. 2009. Matrix factorization tech- niques for recommender systems. Computer 42, 8 (2009), 30–37

  11. [20]

    Lang Lang, Zhenlong Zhu, Xuanye Liu, Jianxin Zhao, Jixing Xu, and Minghui Shan. 2021. Architecture and operation adaptive network for online recommen- dations. In SIGKDD. 3139–3149

  12. [21]

    Honghao Li, Yiwen Zhang, Yi Zhang, Hanwei Li, Lei Sang, and Jieming Zhu

  13. [22]

    Jianxun Lian, Xiaohuan Zhou, Fuzheng Zhang, Zhongxia Chen, Xing Xie, and Guangzhong Sun. 2018. xDeepFM: Combining explicit and implicit feature interactions for recommender systems. In SIGKDD. 1754–1763

  14. [23]

    Kelong Mao, Jieming Zhu, Liangcai Su, Guohao Cai, Yuru Li, and Zhenhua Dong

  15. [24]

    H Brendan McMahan, Gary Holt, David Sculley, Michael Young, Dietmar Ebner, Julian Grady, Lan Nie, Todd Phillips, Eugene Davydov, Daniel Golovin, et al. 2013. Ad click prediction: a view from the trenches. In SIGKDD. ACM, 1222–1230

  16. [25]

    arXiv preprint arXiv:2407.13349 (2024)

    DCNv3: Towards Next Generation Deep Cross Network for CTR Prediction. arXiv preprint arXiv:2407.13349 (2024)

  17. [26]

    Junwei Pan, Jian Xu, Alfonso Lobos Ruiz, Wenliang Zhao, Shengjun Pan, Yu Sun, and Quan Lu. 2018. Field-weighted factorization machines for click-through rate prediction in display advertising. In WWW. 1349–1357

  18. [27]

    Junwei Pan, Wei Xue, Ximei Wang, Haibin Yu, Xun Liu, Shijie Quan, Xueming Qiu, Dapeng Liu, Lei Xiao, and Jie Jiang. 2024. Ad Recommendation in a Collapsed and Entangled World. arXiv preprint arXiv:2403.00793 (2024)

  19. [28]

    Yanru Qu, Han Cai, Kan Ren, Weinan Zhang, Yong Yu, Ying Wen, and Jun Wang

  20. [29]

    Steffen Rendle. 2010. Factorization machines. In ICDM. 995–1000

  21. [30]

    Maxim Naumov, Dheevatsa Mudigere, Hao-Jun Michael Shi, Jianyu Huang, Narayanan Sundaraman, Jongsoo Park, Xiaodong Wang, Udit Gupta, Carole- Jean Wu, Alisson G Azzolini, et al. 2019. Deep learning recommendation model for personalization and recommendation systems.arXiv preprin...

  22. [31]

    Matthew Richardson, Ewa Dominowska, and Robert Ragno. 2007. Predicting clicks: estimating the click-through rate for new ads. In WWW. ACM, 521–530

  23. [32]

    Ruslan Salakhutdinov and Andriy Mnih. 2008. Bayesian probabilistic matrix factorization using Markov chain Monte Carlo. In ICML. 880–887

  24. [33]

    Weiping Song, Chence Shi, Zhiping Xiao, Zhijian Duan, Yewen Xu, Ming Zhang, and Jian Tang. 2019. Autoint: Automatic feature interaction learning via self- attentive neural networks. In CIKM. 1161–1170

  25. [34]

    Product-based neural networks for user response prediction. In ICDM. 1149–1154

  26. [35]

    Yang Sun, Junwei Pan, Alex Zhang, and Aaron Flores. 2021. Fm2: Field-matrixed factorization machines for recommender systems. In WWW. 2828–2837

  27. [36]

    Steffen Rendle, Walid Krichene, Li Zhang, and John Anderson. 2020. Neural collaborative filtering vs. matrix factorization revisited. In RecSys. 240–248

  28. [37]

    Ruoxi Wang, Bin Fu, Gang Fu, and Mingliang Wang. 2017. Deep & cross network for ad click predictions. In ADKDD. 1–7

  29. [38]

    Ruoxi Wang, Rakesh Shivanna, Derek Cheng, Sagar Jain, Dong Lin, Lichan Hong, and Ed Chi. 2021. DCN-V2: Improved deep & cross network and practical lessons for web-scale learning to rank systems. In WWW. 1785–1797

  30. [39]

    Zhiqiang Wang, Qingyun She, and Junlin Zhang. 2021. Masknet: Introducing feature-wise multiplication to CTR ranking models by instance-guided mask. arXiv preprint arXiv:2102.07619 (2021)

  31. [40]

    Liangcai Su, Junwei Pan, Ximei Wang, Xi Xiao, Shijie Quan, Xihua Chen, and Jie Jiang. 2023. STEM: Unleashing the Power of Embeddings for Multi-task Recommendation. arXiv preprint arXiv:2308.13537 (2023)

  32. [41]

    Lanling Xu, Zhen Tian, Gaowei Zhang, Junjie Zhang, Lei Wang, Bowen Zheng, Yifan Li, Jiakai Tang, Zeyu Zhang, Yupeng Hou, Xingyu Pan, Wayne Xin Zhao, Xu Chen, and Ji-Rong Wen. 2023. Towards a More User-Friendly and Easy-to-Use Benchmark Library for Recommender Systems. In SIGIR...

  33. [42]

    Zhen Tian, Ting Bai, Wayne Xin Zhao, Ji-Rong Wen, and Zhao Cao. 2023. Euler- Net: adaptive feature interaction learning via Euler’s formula for CTR prediction. arXiv preprint arXiv:2304.10711 (2023)

  34. [43]

    Weinan Zhang, Tianming Du, and Jun Wang. 2016. Deep learning over multi-field categorical data: A case study on user response prediction. In ECIR. 45–57

  35. [44]

    Wayne Xin Zhao, Yupeng Hou, Xingyu Pan, Chen Yang, Zeyu Zhang, Zihan Lin, Jingsen Zhang, Shuqing Bian, Jiakai Tang, Wenqi Sun, Yushuo Chen, Lanling Xu, Gaowei Zhang, Zhen Tian, Changxin Tian, Shanlei Mu, Xinyan Fan, Xu Chen, and Ji-Rong Wen. 2022. RecBole 2.0: Towards a More U...

  36. [45]

    Wayne Xin Zhao, Shanlei Mu, Yupeng Hou, Zihan Lin, Yushuo Chen, Xingyu Pan, Kaiyuan Li, Yujie Lu, Hui Wang, Changxin Tian, Yingqian Min, Zhichao Feng, Xinyan Fan, Xu Chen, Pengfei Wang, Wendi Ji, Yaliang Li, Xiaoling Wang, and Ji-Rong Wen. 2021. RecBole: Towards a Unified, Com...

  37. [46]

    Jun Xiao, Hao Ye, Xiangnan He, Hanwang Zhang, Fei Wu, and Tat-Seng Chua

  38. [47]

    Jieming Zhu, Quanyu Dai, Liangcai Su, Rong Ma, Jinyang Liu, Guohao Cai, Xi Xiao, and Rui Zhang. 2022. Bars: Towards open benchmarking for recommender systems. In SIGIR. 2912–2923

  39. [48]

    Jieming Zhu, Qinglin Jia, Guohao Cai, Quanyu Dai, Jingjie Li, Zhenhua Dong, Ruiming Tang, and Rui Zhang. 2023. Final: Factorized interaction layer for ctr prediction. In SIGIR. 2006–2010

  40. [49]

    Yi Yang, Baile Xu, Shaofeng Shen, Furao Shen, and Jian Zhao. 2020. Operation- aware neural networks for user response prediction. Neural networks 121 (2020), 161–168

  41. [53]

    Haolin Zhou, Junwei Pan, Xinyi Zhou, Xihua Chen, Jie Jiang, Xiaofeng Gao, and Guihai Chen. 2024. Temporal Interest Network for User Response Prediction. In Companion Proceedings of the ACM on Web Conference 2024 . 413–422

  42. [56]

    Jieming Zhu, Jinyang Liu, Shuai Yang, Qi Zhang, and Xiuqiang He. 2020. Fux- ictr: An open benchmark for click-through rate prediction. arXiv preprint arXiv:2009.05794 (2020). ACM SIGKDD, August 25–29, Barcelona, Spain Yu Kang, Junwei Pan, Jipeng Jin, Shudong Huang, Xiaofeng Ga...

  43. [2016]

    In Proceedings of the 1st workshop on deep learning for recommender systems

    Wide & deep learning for recommender systems. In Proceedings of the 1st workshop on deep learning for recommender systems . 7–10

  44. [2017]

    arXiv preprint arXiv:1708.04617 (2017)

    Attentional factorization machines: Learning the weight of feature interac- tions via attention networks. arXiv preprint arXiv:1708.04617 (2017)

  45. [2021]

    In Recsys

    Reenvisioning the comparison between neural collaborative filtering and matrix factorization. In Recsys. 521–529

  46. [2023]

    FinalMLP: An Enhanced Two-Stream MLP Model for CTR Prediction.arXiv preprint arXiv:2304.00902 (2023)

  47. [2024]

    https://www.iab.com/insights/internet-advertising-revenue-report-2024/

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.