We have overcome the technical bottlenecks of difficult extraction of domain concept relationships and low accuracy of learning service recommendations in graph construction in a multi space fusion environment, and proposed knowledge graph automatic construction and evaluation technology, learning path dynamic planning method, and learning service recommendation technology based on deep matrix decomposition, significantly alleviating the problem of learning confusion caused by massive fragmented resources. Effectively supporting deep mining and precise intelligent guidance for large-scale personalized learning, the specific structure is shown in the following figure.