Authors: Prabhat Agarwal, Minhazul Islam Sk, Nikil Pancha, Kurchi Subhra Hazra, Jiajing Xu, Chuck Rosenberg

Published on: April 25, 2024

Impact Score: 8.0

Arxiv code: Arxiv:2404.16260

Summary

• What is new: OmniSearchSage introduces a unified query embedding for Pinterest search, significantly improving relevance, engagement, and ads CTR.
• Why this is important: The need for a more efficient and accurate search system in Pinterest that understands queries, pins, and products cohesively.
• What the research proposes: A scalable system that jointly learns unified query embeddings alongside pin and product embeddings, using content derived from image captions, historical engagement, and user-curated boards.
• Results: An 8% increase in relevance, 7% increase in engagement, and 5% increase in ads CTR within Pinterest’s search system.

Technical Details

Technological frameworks used: OmniSearchSage system for query understanding and entity representation.

Models used: Generative LLM for enriching text, multitask learning setup for embedding generation.

Data used: Image captions, historical engagement data, user-curated board data.

Potential Impact

This could disrupt the digital advertising and e-commerce marketplaces by offering more relevant search results and, therefore, improving user engagement and ad performance. Companies in these sectors could benefit or need to adapt.

Want to implement this idea in a business?

We have generated a startup concept here: EmbedIQ.