Auto parts recommendation algorithm based on knowledge map

Considering the recommendation for the supply of automotive parts, this paper investigates a recommendation algorithm based on knowledge graphs and graph convolutional networks for automotive component supply. The algorithm performs neighborhood sampling on target nodes, aggregates information from neighboring nodes, and integrates the node's intrinsic features with the aggregated representations to generate higher-order embeddings. These embeddings yield prediction scores for candidate items. Model parameters are optimized through loss minimization, enabling convergence and recommendation generation. The experimental comparisons demonstrate that the proposed model significantly outperforms baseline methods, which include SVD, RippleNet, and KGCN, across accuracy, recall, and F1-score metrics. Specifically, it achieves state-of-the-art performance with 90.5% accuracy, 89.8% recall, and a 90.1% F1-score, further validating its superiority.