Cross-domain Recommendation Model Based on Graph Neural Network and Multi-head Attention Mechanism
ZHU Linfeng
WANG Junhao
Abstract：To address the issue that shared cross-domain recommendation models fail to effectively capture and transfer cross-domain information in data-sparse scenarios,leading to insufficient user preference transfer and decreased recommendation efficiency,a cross-domain recommendation model based on graph neural network and multi-head attention mechanism(GMACDR)was proposed.Firstly,the node to vector(Node2Vec)algorithm was leveraged for graph embedding,where node propagation paths were constructed through random walks,and high-fidelity node representations were learned using the Skip-gram model.Secondly,during the information propagation process within the graph convolutional network,a multi-head attention mechanism was incorporated to dynamically modulate the propagation weights of cross-domain information,effectively capturing intricate user-item interactions.Finally,a dynamic weighting mechanism was employed to aggregate domain-specific features in a weighted manner,refining the feature fusion process and generating more expressive and representative user embeddings.Compared to the best baseline model,the hit rate increased by 4.61%,the normalized discounted cumulative gain increased by 10.47%,and the mean reciprocal rank increased by 9.71%.The research results demonstrated that the model provided more accurate recommendations for users.
Keywords：graph neural networkmulti-head attention mechanismsimilarity matrixcross-domain recommendationdata sparsitydeep learningrecommendation system
Publication Date：2025-09-20
Online Publishing Date：2025-09-24(First online date of this platform, not the publication date of the document)
Pages：7( 418-424 )