The issue of information explosion and information cocoon is becoming increasingly serious, and users need accurate and personalized recommendation methods to enhance their experience. Traditional recommendation methods based on collaborative filtering often face data sparsity problems in actual application scenarios and have difficulty meeting personalized recommendation needs. Inorder to solve the above problems and adapt to the development of modern recommendation tasks, a Knowledge Graph enhanced Multi-task Graph Convolution Recommendation Method (KGCR) is proposed. This method integrates the relationship structure and node attributes in the knowledge graph, learns the potential associations between projects and entities through the knowledge graph embedding module, and applies the Neural Matrix Factorization architecture to design the recommendation module to learn the interaction function between users and items. Finally, the feature transfer mechanism shares feature learning between the two modules. Experimental results in five real recommendation scenarios show that the accuracy of this recommendation method exceeds multiple advanced benchmark models, with an average increase of 6.3%, demonstrating excellent recommendation performance and generalization ability, and helping to solve the problem of data sparsity.