Cross-Domain Interest Representation Learning for Scenario- and Task-Aware Recommendation
Following the modular decomposition of industrial recommenders, STAR-CDR introduces three innovations: a CDI Extractor that enhances sequential modeling with domain indicators and other key features, and a CDI Injector that recalibrates activations and enables domain-/scenario-/task-aware expert routing.