HAMUR: Hyper Adapter for Multi-Domain Recommendation

TL;DR

This work tackles the challenge of Multi-Domain Recommendation by addressing domain bias and rigidity of static parameters. It introduces HAMUR, which couples a domain-specific adapter (plugged into various backbones) with a domain-shared hyper-network that dynamically generates adapter weights at an instance level, using low-rank decomposition for efficiency. The method integrates end-to-end training and demonstrates superior performance across two public CTR datasets and multiple backbone networks, with favorable robustness and scalability. The findings indicate HAMUR can adapt to diverse domain distributions while leveraging shared information, offering a practical path to scalable, multi-domain CTR in real systems.

Abstract

Multi-Domain Recommendation (MDR) has gained significant attention in recent years, which leverages data from multiple domains to enhance their performance concurrently.However, current MDR models are confronted with two limitations. Firstly, the majority of these models adopt an approach that explicitly shares parameters between domains, leading to mutual interference among them. Secondly, due to the distribution differences among domains, the utilization of static parameters in existing methods limits their flexibility to adapt to diverse domains. To address these challenges, we propose a novel model Hyper Adapter for Multi-Domain Recommendation (HAMUR). Specifically, HAMUR consists of two components: (1). Domain-specific adapter, designed as a pluggable module that can be seamlessly integrated into various existing multi-domain backbone models, and (2). Domain-shared hyper-network, which implicitly captures shared information among domains and dynamically generates the parameters for the adapter. We conduct extensive experiments on two public datasets using various backbone networks. The experimental results validate the effectiveness and scalability of the proposed model.

Figures (5)

• Figure 1: Existing Multi-domain sharing methods.
• Figure 2: An illustration of the overall architecture of our proposed model.
• Figure 3: The effect of $K$ on AUC on different datasets.
• Figure 4: The effect of hyper-network dimensions on different datasets.
• Figure 5: Different Backbone Network.