Demystifying Core Ranking in Pinterest Image Search

TL;DR

This paper details Pinterest's practical approach to enhancing image search quality through innovative ranking pipelines, focusing on data, features, and models, validated by extensive offline and online evaluations.

Contribution

It introduces novel research on training data, featurization, and ranking models, and discusses their deployment in Pinterest's image search system.

Findings

Final ranking models show improved search relevance

Offline and online evaluations demonstrate model effectiveness

Efficient deployment of ranking pipelines in production environment

Abstract

Pinterest Image Search Engine helps hundreds of millions of users discover interesting content everyday. This motivates us to improve the image search quality by evolving our ranking techniques. In this work, we share how we practically design and deploy various ranking pipelines into Pinterest image search ecosystem. Specifically, we focus on introducing our novel research and study on three aspects: training data, user/image featurization and ranking models. Extensive offline and online studies compared the performance of different models and demonstrated the efficiency and effectiveness of our final launched ranking models.