SMART: LLM-Augmented Hybrid Retrieval for Dynamic Product Ads

2026-07-25

AI Summary

This work addresses the dual challenge of semantic understanding and retrieval efficiency in dynamic product advertising, which must balance precise remarketing with exploration of new user interests across tens of millions of items. The authors propose SMART, a novel framework that decouples retrieval into two pathways: a rule-based remarketing path executed efficiently over a BM25 index, and an exploration path leveraging large language models (LLMs) to generate queries for dense approximate nearest neighbor (ANN) search. A lightweight adaptive gating mechanism activates the LLM path for only ~10% of high-potential users, preserving nearly all semantic gains while reducing LLM inference costs by 90%. Offline evaluations demonstrate significant relevance improvements, and online A/B tests on Snap’s platform show a 27.6% increase in ad conversion rate.

Abstract

Dynamic Product Ads (DPA) require retrieving relevant items from multi-million product catalogs, balancing two competing objectives: retargeting (re-surfacing known interests) and prospecting (discovering new categories). While Large Language Models (LLMs) capture semantic intent better than traditional embedding models, deploying them at scale introduces prohibitive inference costs and lexical mismatch issues. Through controlled experiments on millions of users, we demonstrate a critical retrieval decomposition: rule-generated queries excel at retargeting on a lexical BM25 index, while LLM-generated queries excel at prospecting on a dense ANN index. Building on this, we propose SMART (SeMantic-aware Adaptive ReTrieval). To manage costs, a lightweight quality gate identifies coverage gaps in initial keyword results, adaptively routing only the ~10% of users who benefit from semantic prospecting to the LLM path. Offline evaluation demonstrates that this gated approach captures the bulk of semantic prospecting gains in Relevance Score while maintaining competitive re-targeting performance at a 90% reduction in LLM costs. Finally, in a 2-week online A/B test at Snap, SMART improved the ad conversion rate by +27.6% over a strong embedding-based baseline.

Problem

• Dynamic Product Ads
• Retargeting
• Prospecting
• Large Language Models
• Hybrid Retrieval

Innovation

• hybrid retrieval
• large language models
• dynamic product ads
• semantic prospecting
• cost-efficient inference