Identity preserving adversarial training method and device based on graph representation learning, and medium

Abstract

The invention provides an identity preserving adversarial training method and device based on graph representation learning, and a medium. The method comprises the following steps: acquiring graph data of a training scene, defining each node of the graph data as an original sample for representing the training scene, and defining sample identity information of the original sample; generating an adversarial sample corresponding to each original sample; by adding an identity keeping constraint to the adversarial sample, keeping sample identity information of an original sample for the adversarial sample; taking the adversarial sample as a first input variable, inputting the adversarial sample into the initial graph representation learning model, and executing identity preserving adversarial training; and updating the initial graph representation learning model to obtain a target graph representation learning model, and predicting the output of the original sample under different graph mining tasks in the training scene by using the target graph representation learning model. According to the method, the adversarial sample and the original sample keep the same sample identity information, the precision of graph representation learning in graph structure data analysis is improved, and the method has certain universality.

Description

technical field [0001] The present invention relates to the technical field of graph data mining, in particular to a method, device and medium for identity preservation confrontation training based on graph representation learning. Background technique [0002] Graph representation learning has become a popular research area for analyzing graph-structured data. At the software level, graph representation learning aims to learn an encoding function that takes full advantage of graph data to transform graph data with complex structures into dense representations in low-dimensional spaces that preserve diverse graph properties and structural features. At present, graph representation learning methods are widely used in various graph mining tasks such as node classification, anomaly detection, edge prediction, and label recommendation. At the same time, it has brought breakthroughs to a large number of application problems in real life. For example, in the product recommendatio...

AI Technical Summary

Problems solved by technology

However, since this method does not guarantee the quality of the adversarial samples, it is easy to introduce error information into the adversarial samples and damage the performance of the graph representation learning model, which affects the nodes in actual training scenarios such as product recommendation scenarios, drug target prediction, and financial risk control scenarios. Accuracy of graph structured data analysis in various graph mining tasks such as classification, anomaly detection, edge prediction, label recommendation, etc.

Images