Intelligent library recommendation based on GNN and attention networks

Abstract

In the era of digital transformation, libraries are faced with the challenge of finding a balance between massive resources and personalized user needs. Existing recommendation models often cannot handle complex user-item interaction networks and are difficult to fully explore the deep connection between user preferences and books. To address this problem, the study proposes a novel intelligent recommendation model that uniquely integrates the structural representation ability of Graph Neural Networks with the dynamic feature extraction ability of attention mechanisms. This innovative combination enables the simultaneous capture of complex user-item topological relationships and adaptive importance weighting of heterogeneous features, which has not been extensively explored in library recommendation systems. The model uses a multi-head attention mechanism for adaptive feature selection and combines residual connections to ensure the effective flow of information in the deep network. Experimental results show that on the Amazon book dataset, the proposed fusion model outperforms traditional deep learning methods and improves the recommendation accuracy by more than 15%. In the actual deployment of a university library, the recommendation coverage of the fusion model reached 89.2%, the user satisfaction score was 4.52 points (out of 5 points), and the recommendation accuracy in multiple subject areas remained above 85%. These results show that the fusion model effectively reveals deep relationships, solves the challenges of modeling dynamic user interests and diverse book features, and provides reliable technical support for the intelligent and personalized reading recommendation system of modern libraries.