Multi-Head Attention and Knowledge Graph Based Dual Target Graph Collaborative Filtering Network

Recently, cross-domain collaborative filtering (CDCF) has been widely used to solve the data sparsity problem in recommendation systems. Therein, dual-target cross-domain recommendation becomes a research hotspot, which aims to improve the recommendation performance of both target and source domains. Most existing approaches tend to use fixed weights or self-attention in a single representation space for the bi-directional inter-domain transfer of the user representation. However, a single representation space leads to limited representation capability, which makes the transfer of the user representation inaccurate. Also, the existing approaches ignore the representation enhancement of items. In this paper, Multi-head Attention and Knowledge Graph Based Dual Target Graph Collaborative Filtering Network (MAKG-DTGCF) is proposed. Based on the graph collaborative filtering model, it uses the multi-head attention based bi-directional transfer module to realize adaptive transfer and fusion of user features in multiple representation subspaces, which can make the transfer of user representation more accurate. Meanwhile, we also enhance the item representation by aligning the item embedding in the user-item heterogeneous graph with the knowledge embedding of the item in the knowledge graph. Experimental results on three real datasets show that the proposed MAKG-DTGCF model significantly outperforms the state-of-the-art models in terms of HR and NDCG.

Automated summary

Recently, cross-domain collaborative filtering (CDCF) has gained popularity in addressing the data sparsity issue in recommendation systems. This paper proposes a Multi-head Attention and Knowledge Graph Based Dual Target Graph Collaborative Filtering Network (MAKG-DTGCF) to improve the recommendation performance of both target and source domains. The MAKG-DTGCF model utilizes multi-head attention for adaptive transfer and fusion of user features in multiple representation subspaces, and enhances item representation through alignment with knowledge graphs. Experimental results demonstrate that the MAKG-DTGCF model outperforms state-of-the-art models in HR and NDCG metrics.