Traditional recommendation systems, particularly those based on collaborative filtering, have exhibited robust performance by harnessing the advantages of big data. In these systems, the essence lies in identifying users whose interests closely align with a target user and subsequently recommending items that align with their preferences, showcasing commendable effectiveness in delivering personalized suggestions. Traditional recommender algorithms based on Graph Neural Networks (GNNs) face limitations as they can only handle regular topological graphs composed of a single type of node. However, in contemporary networks, data is often not exclusively comprised of a singular node type. Additionally, conventional GNNs are constrained to incorporating only first-order neighbor features of nodes, lacking the ability to capture deeper structural relationships within the network. Consequently, when dealing with sparse datasets where nodes have very few neighbors, the recommendation quality of algorithms based on traditional GNNs significantly diminishes. To address the aforementioned limitations, this paper proposes a deep recommendation model that combines Graph Neural Networks (GNNs) with heterogeneous networks. By integrating GNNs with heterogeneous network structures, the aim is to overcome the challenges posed by single-node type limitations and the inability to capture deeper structural relationships in traditional recommendation algorithms.