Star+: A New Multi-Domain Model for CTR Prediction

Star+ is a novel multi-domain click-through rate (CTR) prediction model inspired by the Star model. It addresses the challenges of domain-specific data distributions and interdomain relationships in multi-domain environments. Star+ enhances interactions between shared and domain-specific information using fusion strategies such as add, adaptive add, concatenation, and gating, and evaluates normalization techniques like layer, batch, and partition normalization. Extensive experiments demonstrate significant improvements in accuracy and efficiency, advancing scalable and adaptive recommendation systems.