Skip to main content
Back

SPHERE Uses LLMs to Break Recommendation Silos Between Unconnected Platforms

SPHERE uses LLMs to build behavioral personas that transfer recommendations across platforms with zero shared users or items

Paper, Full Access

Every time you rate a book on Goodreads, that signal stays locked inside Goodreads. Steam has no idea you devoured sci-fi novels all winter, even though that pattern might predict your next game purchase perfectly. This is the information silo problem, and it is the norm rather than the exception across independent digital platforms. A new paper introduces SPHERE (Semantic Personas for Heterogeneous cross-domain Recommendation), a system that breaks those silos using LLMs to transfer behavioral knowledge between platforms that share absolutely nothing -- no common users, no common items, no common structure.