HyperGate: Hierarchical Perceptive Gating Network for Multi-domain Multi-task Recommendation

The growing prevalence of interactive pages in online services presents novel challenges for industrial recommendation systems, particularly in the realms of multi-domain and multi-task recommendations.These systems must not only model commonality across various domains but also account for distinctions between domains and tasks.However, the lack of explicit semantic information regarding domains and tasks complicates the process.In this paper, we propose a novel parameter-sharing model for multi-domain multi-task recommendations, termed the Hierarchical Perceptive Gating Network (pronounced as HyperGate).Drawing inspiration from Collective Behavior Theory, our approach first augments semantic information for domains and tasks using contrastive learning, which then serves as perceptive control for the subsequent hierarchical gating network.The hierarchical gating network consists of three key components, including a domainperceptive embedding gate, a domain-perceptive biasing gate, and a dual-perceptive fusion gate, thereby constructing a domain-and task-perceptive parameter-sharing network from the bottom up.Experiments conducted on two real-world datasets validate