Graph convolution recommendation system based on knowledge graph

In recommendation systems, nodes in the network can be represented as vertices on a graph, and the connections between these nodes are represented as edges. By using this viewpoint, a network can be converted into an all-inclusive graph that includes many items and their complex interrelationships. Furthermore, because of their remarkable ability to extract both structural and semantic information from graphs, Graph Convolutional Neural networks (GCNs) have attracted a lot of attention lately. However, many of the product models in the field of graph convolution-based recommendation algorithms ignore the rich multisource auxiliary information buried within the network by simplifying it to a bipartite graph for convolution operations. The precision and variety of suggestion results may be jeopardized by this simplicity. This article presents a graph convolutional recommendation model that smoothly incorporates knowledge graphs in order to overcome this restriction. In particular, by outlining the items and their connections inside the network, the model initially creates a knowledge graph. It then randomly navigates the knowledge graph using meta-paths to find possible semantic relationships between nodes. The distinguishing characteristics of these nodes are subsequently extracted via an attention method. The node features obtained from the bipartite graph convolution are then fused with these extracted features. Predictive scores are then produced by projecting several data points onto a single dimension using a multilayer perceptron. The effectiveness of our suggested method has been confirmed by experimental findings on a variety of datasets.