Attribute network embedding method based on personalized relation sorting

The invention relates to an attribute network embedding method based on personalized relation sorting, belonging to the representation learning field of machine learning. The purpose of this method isto represent the attribute network in vector form, so that the useful knowledge in mining network can be analyzed better. Firstly, the similarity matrices of attributes and topology between nodes inattribute network are calculated, and the similarities between nodes in these two aspects are sorted by using two thresholds. Then we synthesize the two sorts of relation to determine the result of the comprehensive relation sorting among the nodes, and give the objective function to be optimized. Finally, using the expectation maximization algorithm and the gradient rise principle, by iterativelyupdating the threshold and the embedding vector, the objective function is optimized, and the embedding vector of all nodes is obtained. The embedding vector generated by this method is not only easyto analyze and deal with, b