Convolutional network feature learning method based on graph data model

The invention relates to the technical field of machine learning, in particular to a convolutional network feature learning method based on a graph data model, which comprises the following steps of: representing feature learning in a network as a maximum likelihood optimization problem, and proposing two standard hypotheses based on the maximum likelihood optimization problem, simplifying the maximum likelihood optimization problem based on two standard hypotheses, and searching the network neighborhood by using a random walk strategy; setting two parameters b and d to guide the random walk strategy; and optimizing the features by using stochastic gradient descent to obtain the newest feature representation, and finally applying the newest feature representation to the graph convolutional network model to carry out downstream tasks. According to the method and the device, the problems that in the prior art, due to the fact that the method depends on a fixed network neighborhood concept, the method and the de