Reproduce “OmniSearchSage: Multi-Task Multi-Entity Embeddings for Pinterest Search” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Prabhat Agarwal, Minhazul Islam Sk, Nikil Pancha, Kurchi Subhra Hazra, Jiajing Xu, Chuck Rosenberg
Pinterest's OmniSearchSage system unifies an embedding framework across user queries, pins, and products using a multi-task learning approach. This system led to over 8% improvement in search relevance, over 7% increase in user engagement, and over 5% improvement in ad click-through rates across Pinterest's search platform.
Key insights
- Unified query embeddings that function across multiple entity types (queries, pins, products) are highly effective.
- Text enrichment, particularly through synthetic captions and engagement queries, is critical for robust content representation.
- Compatibility with pre-existing embedding systems like PinSage and ItemSage is essential for practical system integration and deployment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.