Apple Augments App Store Ranking With LLMs
Apple researchers published a study, 'Scaling Search Relevance: Augmenting App Store Ranking with LLM-Generated Judgments,' and ran a worldwide A/B test to evaluate LLM-generated textual relevance labels. They fine-tuned a 3-billion-parameter LLM, generated millions of labels, and retrained the App Store ranker. The LLM-augmented model produced a statistically significant +0.24% lift in search-session conversion rate across 89% of storefronts, implying millions more downloads.
Key Points
- 1Generated millions of textual relevance labels using a fine-tuned 3B-parameter LLM
- 2Demonstrated significance: LLM labels address scarcity of human textual relevance judgments
- 3Translated to a +0.24% worldwide conversion lift, implying millions more downloads
Scoring Rationale
Credible production A/B test showing measurable uplift; limited novelty beyond established LLM-based labeling approaches today.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Logistics & Shipping problems
