Progressive heterogeneous network robust representation learning method without element path supervision

The invention discloses a progressive heterogeneous network robust representation learning method without meta path supervision, and the method comprises the steps: constructing a heterogeneous network robust representation learning model, and eliminating the dependence of a conventional model on a meta path through employing two unsupervised independent learning processes: structural feature learning and semantic feature learning. Meanwhile, random sampling and adversarial learning mechanisms are integrated into two learning processes to eliminate the influence of noise and improve the robustness of the model, and the method specifically comprises the following steps: 1) constructing a type-level isomorphic representation strategy as a rough embedding stage, and capturing a rough representation vector containing structural features by using a scheme of type perception sampling + intra-class feature learning + inter-class feature aggregation; 2) constructing a relationship-level heterogeneous representation s