With the rapid development of Internet technology and the comprehensive popularity of Internet applications, online activities have gradually become an indispensable part of people's daily life. Internet activities are becoming an indispensable part of people's daily life. How to help users quickly find valuable information from the huge amount of data has become a great challenge for the Internet industry. The traditional recommendation system based on collaborative filtering can take advantage of big data because it can well find users with the most similar interests and recommend the items they are interested in, showing a good recommendation effect and has been the main direction of recommendation algorithm research. In recommender systems, most of the information has a graph structure, and the graph neural network GNN technique can capture the relevance of graphs through message passing between graph nodes, so GNNs are often used to generate embedded representations of users/items. However, traditional GNN-based recommendation algorithms are only able to handle regular topological graphs composed of a single type of node, while the data in the current network is not composed of only a single type of node. the main content of this paper is the GNN-based knowledge graph recommendation model. The model can effectively capture the potential interest of users through the user relationship graph for the recommendation. A joint multi-feature representation method is proposed. For user-item interaction, the representation of node features in the case of sparse network connectivity is enhanced by using multi-order topological structure information in information networks. The model in this paper was verified to perform significantly better than the baseline approach by conducting experiments on public datasets.