A Dual-Channel Heterogeneous Hypergraph Convolutional Network for Dual-Target Cross-Domain Recommendation

Abstract

Cross-domain recommendation (CDR), which aims to alleviate the data sparsity problem in a single domain by integrating complementary data from multiple domains, has become a practical and challenging research direction. Although achieving promising performance, we highlight two problems in existing CDR methods. 1) The representation ability of existing ID-based item embedding is limited. 2) Knowledge transferability across different domains is often insufficient. To solve these problems, we propose a new cross-domain recommendation method, termed D ual-Channel H eterogeneous H yper G raph C onvolutional N etwork (DHHGCN), which primarily consists of two components: the intra-domain channel layer and the inter-domain channel layer. Concretely, within the intra-domain context, we introduce additional item features and build heterogeneous hypergraphs to model fine-grained high-order correlations, resulting in high-quality user and item representations. In terms of the inter-domain, based on designed similarity matrices, we construct hypergraphs and guide the network to learn the relationships via hypergraph convolution, effectively transferring cross-domain knowledge. Last, an element-wise gating mechanism is designed to integrate domain-specific knowledge with shared cross-domain knowledge, enabling dual-target recommendations. Extensive experiments demonstrate the superiority and effectiveness. Our code is available on GitHub ( https://github.com/idleslob/DHHGCN ).