OmniSearchSage: Multi-Task Multi-Entity Embeddings for Pinterest Search

TL;DR

OmniSearchSage introduces a unified multi-task embedding system for Pinterest search, significantly enhancing relevance, engagement, and ad CTR through improved content understanding and real-time scalable deployment.

Contribution

The paper presents a novel multi-task learning framework that jointly learns query, pin, and product embeddings, leveraging diverse text sources and enabling scalable, real-time Pinterest search improvements.

Findings

>8% relevance improvement

>7% engagement increase

>5% ads CTR boost

Abstract

In this paper, we present OmniSearchSage, a versatile and scalable system for understanding search queries, pins, and products for Pinterest search. We jointly learn a unified query embedding coupled with pin and product embeddings, leading to an improvement of $>8\%$ relevance, $>7\%$ engagement, and $>5\%$ ads CTR in Pinterest's production search system. The main contributors to these gains are improved content understanding, better multi-task learning, and real-time serving. We enrich our entity representations using diverse text derived from image captions from a generative LLM, historical engagement, and user-curated boards. Our multitask learning setup produces a single search query embedding in the same space as pin and product embeddings and compatible with pre-existing pin and product embeddings. We show the value of each feature through ablation studies, and show the effectiveness of a unified model compared to standalone counterparts. Finally, we share how these embeddings have been deployed across the Pinterest search stack, from retrieval to ranking, scaling to serve $300k$ requests per second at low latency. Our implementation of this work is available at https://github.com/pinterest/atg-research/tree/main/omnisearchsage.