mlsanityData to 14 Oct 2026

Warmer for Less: A Cost-Efficient Strategy for Cold-Start Recommendations at Pinterest

arXiv 2512.17277 ↗First posted 19 Dec 2025cs.IR cs.LGcited by 2

Not on the map. Why?

Abstract (arXiv)

Pinterest is a leading visual discovery platform where recommender systems (RecSys) are key to delivering relevant, engaging, and fresh content to our users. In this paper, we study the problem of improving RecSys model predictions for cold-start (CS) items, which appear infrequently in the training data. Although this problem is well-studied in academia, few studies have addressed its root causes effectively at the scale of a platform like Pinterest. By investigating live traffic data, we identified several challenges of the CS problem and developed a corresponding solution for each: First, industrial-scale RecSys models must operate under tight computational constraints. Since CS items are a minority, any related improvements must be highly cost-efficient. To address this, our solutions were designed to be lightweight, collectively increasing the total parameters by only 5%. Second, CS items are represented only by non-historical (e.g., content or attribute) features, which models often treat as less important. To elevate their significance, we introduce a residual connection for the non-historical features. Third, CS items tend to receive lower prediction scores compared to non-CS items, reducing their likelihood of being surfaced. We mitigate this by incorporating a score regularization term into the model. Fourth, the labels associated with CS items are sparse, making it difficult for the model to learn from them. We apply the manifold mixup technique to address this data sparsity. Implemented together, our methods increased fresh content engagement at Pinterest by 10% without negatively impacting overall engagement and cost, and have been deployed to serve over 570 million users on Pinterest.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 30 of its 71 references that are in the map. First posted on arXiv 19 Dec 2025.

Builds on

30 of 71 references.

References · 30

  1. 2017Deeper, Broader and Artier Domain Generalizationcited by 986
  2. 2017Learning to Generalize: Meta-Learning for Domain Generalizationcited by 763
  3. 2013Domain Generalization via Invariant Feature Representationcited by 623
  4. 2019Domain Generalization via Model-Agnostic Learning of Semantic Featurescited by 407
  5. 2020DCN V2: Improved Deep & Cross Network and Practical Lessons for Web-scale Learning to Rank Systemscited by 327
  6. 2017Unified Deep Supervised Domain Adaptation and Generalizationcited by 323
  7. 2020Deep Domain-Adversarial Image Generation for Domain Generalisationcited by 231
  8. 2016Unbiased Learning-to-Rank with Biased Feedbackcited by 157
  9. 2020Learning from Extrinsic and Intrinsic Supervisions for Domain Generalizationcited by 108
  10. 2020Model-Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender Systemcited by 94
  11. 2019Managing Popularity Bias in Recommender Systems with Personalized Re-rankingcited by 82
  12. 2021Contrastive Learning for Cold-Start Recommendationcited by 79
18 more
  1. 2020Self-Supervised Hypergraph Convolutional Networks for Session-based Recommendationcited by 75
  2. 2019Warm Up Cold-start Advertisements: Improving CTR Predictions via Learning to Learn ID Embeddingscited by 66
  3. 2021Contrastive Self-supervised Sequential Recommendation with Robust Augmentationcited by 56
  4. 2021Deconfounded Recommendation for Alleviating Bias Amplificationcited by 51
  5. 2021Learning to Warm Up Cold Item Embeddings for Cold-start Recommendation with Meta Scaling and Shifting Networkscited by 37
  6. 2023TransAct: Transformer-based Realtime User Action Model for Recommendation at Pinterestcited by 35
  7. 2020ESAM: Discriminative Domain Adaptation with Non-Displayed Items to Improve Long-Tail Performancecited by 30
  8. 2022Incorporating Bias-aware Margins into Contrastive Loss for Collaborative Filteringcited by 29
  9. 2017Related Pins at Pinterest: The Evolution of a Real-World Recommender Systemcited by 24
  10. 2022IntTower: the Next Generation of Two-Tower Model for Pre-Ranking Systemcited by 18
  11. 2024OmniSearchSage: Multi-Task Multi-Entity Embeddings for Pinterest Searchcited by 13
  12. 2024A Multimodal Single-Branch Embedding Network for Recommendation in Cold-Start and Missing Modality Scenarioscited by 13
  13. 2020Connecting User and Item Perspectives in Popularity Debiasing for Collaborative Recommendationcited by 12
  14. 2020Learning Transferrable Parameters for Long-tailed Sequential User Behavior Modelingcited by 12
  15. 2023Cold & Warm Net: Addressing Cold-Start Users in Recommender Systemscited by 7
  16. 2022Latent User Intent Modeling for Sequential Recommenderscited by 5
  17. 2024Scene-wise Adaptive Network for Dynamic Cold-start Scenes Optimization in CTR Predictioncited by 4
  18. 2023Bootstrapping Contrastive Learning Enhanced Music Cold-Start Matchingcited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026PinEqualizer: Full Funnel Content Exploration and Debiasing System at Pinterestcited by 1
  2. 2026FORTE: A Production Framework for Feature Evaluation and Deprecation in Industrial Recommender Systemscited by 0