In order to solve the problems in the field of music recommendation, such as the difficulty of multi-source heterogeneous data integration, insufficient semantic association mining and insufficient accuracy of personalized recommendation, a recommendation algorithm based on knowledge mapping and deep learning is proposed. Through dynamic crawler technology and UIE intelligent extraction framework, a multi-dimensional music data system is constructed, and the precise construction of knowledge map is realized by using the dual integration strategy of "semantic computing + word form matching". The TransR model is introduced to embed the deep semantics of knowledge map, and the "content-behavior" dual-channel recommendation model is constructed based on the user's historical behavior characteristics. The experimental results show that the proposed algorithm is significantly superior to the existing recommendation algorithms in the key indicators of recommendation accuracy, ranking rationality and user satisfaction, and the research results not only provide a new technical path for music recommendation, but also verify the unique role of knowledge mapping in improving the interpretability of recommendation systems.