PinRec: Outcome-Conditioned, Multi-Token Generative Retrieval for Industry-Scale Recommendation Systems

2025-04-09

AI Summary

Generative retrieval faces scalability limitations and inflexible multi-objective optimization in industrial-scale recommendation systems. To address these challenges, we propose a large-scale outcome-conditioned generative retrieval framework. Our approach introduces an outcome-conditioning mechanism—enabling explicit, tunable trade-offs among key metrics such as save rate and click-through rate—and designs a multi-token parallel decoding paradigm that simultaneously preserves output diversity and improves throughput. Furthermore, it synergistically integrates Transformer-based sequence modeling with production-grade indexing optimizations. Deployed across Pinterest’s full-scale platform, the method achieves significant gains in user engagement: exploration diversity increases by 12.7%, and the system demonstrates robust performance at the billion-user scale—the first empirical validation of generative retrieval’s feasibility and effectiveness in industrial settings. This work establishes a new paradigm for scalable, controllable, and multi-objective-coordinated generative retrieval in production recommendation systems.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Search and Optimization: Learning to Search
• Natural Language Processing: Generation

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs

Abstract

Generative retrieval methods utilize generative sequential modeling techniques, such as transformers, to generate candidate items for recommender systems. These methods have demonstrated promising results in academic benchmarks, surpassing traditional retrieval models like two-tower architectures. However, current generative retrieval methods lack the scalability required for industrial recommender systems, and they are insufficiently flexible to satisfy the multiple metric requirements of modern systems. This paper introduces PinRec, a novel generative retrieval model developed for applications at Pinterest. PinRec utilizes outcome-conditioned generation, enabling modelers to specify how to balance various outcome metrics, such as the number of saves and clicks, to effectively align with business goals and user exploration. Additionally, PinRec incorporates multi-token generation to enhance output diversity while optimizing generation. Our experiments demonstrate that PinRec can successfully balance performance, diversity, and efficiency, delivering a significant positive impact to users using generative models. This paper marks a significant milestone in generative retrieval, as it presents, to our knowledge, the first rigorous study on implementing generative retrieval at the scale of Pinterest.

Problem

• Addressing scalability issues in generative retrieval for industry-scale recommender systems
• Enhancing flexibility to meet multiple metric requirements in modern recommender systems
• Balancing performance, diversity, and efficiency in generative retrieval models

Innovation

• Outcome-conditioned generation for metric balancing
• Multi-token generation enhancing output diversity
• Scalable transformer-based generative retrieval model