Technology Category
Machine Learning: Hardware-aware MLComputer Vision: Learning & Optimization for CVSearch and Optimization: Learning to Search
ML-DCN: Masked Low-Rank Deep Crossing Network Towards Scalable Ads Click-through Rate Prediction at Pinterest
🤖 AI Summary
This work addresses the challenge of improving click-through rate (CTR) prediction under stringent latency and computational constraints by efficiently modeling high-order feature interactions. We propose ML-DCN, a novel architecture that integrates the strengths of DCNv2 and MaskNet through an instance-conditioned masking mechanism, which dynamically selects and amplifies critical interaction directions within low-rank cross layers. This approach substantially enhances model expressiveness under a fixed FLOPs budget without incurring additional inference overhead. Evaluated on Pinterest’s advertising dataset, ML-DCN achieves superior AUC compared to both DCNv2 and MaskNet at identical computational costs. Online A/B tests further demonstrate significant improvements in both CTR and click quality, leading to full-scale deployment without increasing serving costs.
Technology Category
Application Category
📝 Abstract
Deep learning recommendation systems rely on feature interaction modules to model complex user-item relationships across sparse categorical and dense features. In large-scale ad ranking, increasing model capacity is a promising path to improving both predictive performance and business outcomes, yet production serving budgets impose strict constraints on latency and FLOPs. This creates a central tension: we want interaction modules that both scale effectively with additional compute and remain compute-efficient at serving time. In this work, we study how to scale feature interaction modules under a fixed serving budget. We find that naively scaling DCNv2 and MaskNet, despite their widespread adoption in industry, yields rapidly diminishing offline gains in the Pinterest ads ranking system. To overcome aforementioned limitations, we propose ML-DCN, an interaction module that integrates an instance-conditioned mask into a low-rank crossing layer, enabling per-example selection and amplification of salient interaction directions while maintaining efficient computation. This novel architecture combines the strengths of DCNv2 and MaskNet, scales efficiently with increased compute, and achieves state-of-the-art performance. Experiments on a large internal Pinterest ads dataset show that ML-DCN achieves higher AUC than DCNv2, MaskNet, and recent scaling-oriented alternatives at matched FLOPs, and it scales more favorably overall as compute increases, exhibiting a stronger AUC-FLOPs trade-off. Finally, online A/B tests demonstrate statistically significant improvements in key ads metrics (including CTR and click-quality measures) and ML-DCN has been deployed in the production system with neutral serving cost.
Problem
Research questions and friction points this paper is trying to address.
click-through rate prediction
feature interaction
model scaling
serving budget
ad ranking
Innovation
Methods, ideas, or system contributions that make the work stand out.
Masked Low-Rank
Feature Interaction
Scalable CTR Prediction
Instance-Conditioned Mask
Deep Crossing Network
🔎 Similar Papers
No similar papers found.
