# Not All Candidates are Created Equal: A Heterogeneity-Aware Approach to Pre-ranking in Recommender Systems

arXiv: https://arxiv.org/abs/2603.03770 · HTML: https://mlsanity.com/paper/2603.03770/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 19 of its 31 references that are in the map.
- First posted on arXiv 4 Mar 2026.

## Builds on (most cited first, 19 of 19)

- [Search-based User Interest Modeling with Lifelong Sequential Behavior Data for Click-Through Rate Prediction](https://mlsanity.com/paper/2006.05639/) (2020)
- [Practice on Long Sequential User Behavior Modeling for Click-Through Rate Prediction](https://mlsanity.com/paper/1905.09248/) (2019)
- [Learning Tree-based Deep Model for Recommender Systems](https://mlsanity.com/paper/1801.02294/) (2018)
- [KuaiRec: A Fully-observed Dataset and Insights for Evaluating Recommender Systems](https://mlsanity.com/paper/2202.10842/) (2022)
- [Yelp Dataset Challenge: Review Rating Prediction](https://mlsanity.com/paper/1605.05362/) (2016)
- [Contrastive Learning for Debiased Candidate Generation in Large-Scale Recommender Systems](https://mlsanity.com/paper/2005.12964/) (2020)
- [COLD: Towards the Next Generation of Pre-Ranking System](https://mlsanity.com/paper/2007.16122/) (2020)
- [Cascade Ranking for Operational E-commerce Search](https://mlsanity.com/paper/1706.02093/) (2017)
- [Privileged Features Distillation at Taobao Recommendations](https://mlsanity.com/paper/1907.05171/) (2019)
- [IntTower: the Next Generation of Two-Tower Model for Pre-Ranking System](https://mlsanity.com/paper/2210.09890/) (2022)
- [Semantic Modelling with Long-Short-Term Memory for Information Retrieval](https://mlsanity.com/paper/1412.6629/) (2014)
- [Towards a Better Tradeoff between Effectiveness and Efficiency in Pre-Ranking: A Learnable Feature Selection based Approach](https://mlsanity.com/paper/2105.07706/) (2021)
- [Sequential Evaluation and Generation Framework for Combinatorial Recommender System](https://mlsanity.com/paper/1902.00245/) (2019)
- [COPR: Consistency-Oriented Pre-Ranking for Online Advertising](https://mlsanity.com/paper/2306.03516/) (2023)
- [Rethinking Large-scale Pre-ranking System: Entire-chain Cross-domain Models](https://mlsanity.com/paper/2310.08039/) (2023)
- [AutoFAS: Automatic Feature and Architecture Selection for Pre-Ranking System](https://mlsanity.com/paper/2205.09394/) (2022)
- [Rethinking the Role of Pre-ranking in Large-scale E-Commerce Searching System](https://mlsanity.com/paper/2305.13647/) (2023)
- [A Hybrid Cross-Stage Coordination Pre-ranking Model for Online Recommendation Systems](https://mlsanity.com/paper/2502.10284/) (2025)
- [Both Efficiency and Effectiveness! A Large Scale Pre-ranking Framework in Search System](https://mlsanity.com/paper/2304.02434/) (2023)

## Abstract (arXiv)

Most large-scale recommender systems follow a multi-stage cascade of retrieval, pre-ranking, ranking, and re-ranking. A key challenge at the pre-ranking stage arises from the heterogeneity of training instances sampled from coarse-grained retrieval results, fine-grained ranking signals, and exposure feedback. Our analysis reveals that prevailing pre-ranking methods, which indiscriminately mix heterogeneous samples, suffer from gradient conflicts: hard samples dominate training while easy ones remain underutilized, leading to suboptimal performance. We further show that the common practice of uniformly scaling model complexity across all samples is inefficient, as it overspends computation on easy cases and slows training without proportional gains. To address these limitations, this paper presents Heterogeneity-Aware Adaptive Pre-ranking (HAP), a unified framework that mitigates gradient conflicts through conflict-sensitive sampling coupled with tailored loss design, while adaptively allocating computational budgets across candidates. Specifically, HAP disentangles easy and hard samples, directing each subset along dedicated optimization paths. Building on this separation, it first applies lightweight models to all candidates for efficient coverage, and further engages stronger models on the hard ones, maintaining accuracy while reducing cost. This approach not only improves pre-ranking effectiveness but also provides a practical perspective on scaling strategies in industrial recommender systems. HAP has been deployed in the Toutiao production system for 9 months, yielding up to 0.4% improvement in user app usage duration and 0.05% in active days, without additional computational cost. We also release a large-scale industrial hybrid-sample dataset to enable the systematic study of source-driven candidate heterogeneity in pre-ranking.