Research on Recommendation Algorithm of Graph Neural Network Based on Multi-Mode Fusion

This paper introduces a graph neural network recommendation algorithm based on multi-modal fusion, aimed at enhancing recommendation system performance by integrating image features with graph structure features. The central approach of this research involves using Convolutional Neural Networks (CNNs) to extract image features, Graph Convolutional Networks (GCNs) to capture graph structure features, and employing a multi-head attention mechanism to fuse these two types of features. The model processes image data using CNNs to generate high-dimensional feature vectors, while GCNs handle graph structure data and extract the relational information between nodes to form graph structure features. Subsequently, the multi-head attention mechanism is applied to dynamically assign weights to both image and graph structure features. By calculating query, key, and value representations for each mode and using multiple attention heads to capture various subspaces in parallel, the model learns the intricate relationships between these two modalities. Experimental results demonstrate that the proposed algorithm achieves strong performance on multiple public datasets, particularly for tasks involving image and graph structure data. The model effectively leverages the complementarity of different modalities, leading to significant improvements in recommendation performance.