In graph-based recommendation, the calculation and storage of resource allocation matrix take up a lot of overhead, in order to improve efficiency of algorithm. Path Integration-Based Multiple Layers Recommendation(PIML) was proposed. In the algorithm, path integration-based shunt strategy was used to dynamically allocate resources, and timestamp and ratings were added to the graph. On the other hand, demographic information and item content information were added to the multiple layers structure, considering a variety of factors, it gave more comprehensive recommendation. Experimental results show that the algorithm improves the accuracy without increasing time cost; and makes recommendation more comprehensive and can give real-time recommendation.