Graph depth clustering method based on double-correlation reduction

Abstract

The invention relates to a graph depth clustering method and device based on double-correlation reduction, computer equipment and a storage medium. The method comprises the following steps: acquiring a to-be-clustered original target relation graph; performing data enhancement on the original target relation graph based on the feature similarity and graph diffusion to obtain a classification graph adjacency matrix and a diffusion matrix, obtaining a first augmented graph according to the classification graph adjacency matrix and Gaussian noise, and obtaining a second augmented graph according to the diffusion matrix and the Gaussian noise; respectively inputting the feature maps of the first augmented map and the second augmented map into a pre-trained double-correlation reduction network to carry out correlation reduction of a sample level and correlation reduction of a feature level, and obtaining a double-correlation reduction feature map; and performing graph clustering on the double-correlation reduced feature graph to obtain a target relation clustering graph. By adopting the method, collapse can be avoided, and the clustering performance is improved.

Description

technical field [0001] The present application relates to the technical field of deep learning, and in particular, to a graph depth clustering method based on double correlation reduction. Background technique [0002] Deep graph clustering is the training of a deep neural network to learn embedded representations of nodes to classify nodes into distinct clusters without human annotation. Graph convolutional networks (GCNs) have recently achieved promising performance in many graph clustering applications such as social networks and recommender systems due to their powerful graph information capture capabilities. Benefiting from the powerful graph information extraction ability, GNN has achieved promising results in graph depth clustering, but the problem of over-smoothing remains unsolved. Since then, Structured Deep Clustering Network (SDCN) / Structural Deep Clustering Network with Q Distribution (SDCN_Q) and Deep Fusion Clustering Network (DFCN) proposed to jointly train ...

AI Technical Summary

Problems solved by technology

[0003] However, traditional clustering algorithms based on graph convolutional neural networks usually suffer from the problem of representation collapse, that is, they tend to map different categories of nodes to similar representations during the sample embedding process, for example, GAE and MVGRL (Multi-ViewGraph Representation Learning), GAE is a classic graph convolutional network, and MVGRL is an algorithm enhanced by using a contrastive learning strategy. By introducing a positive and negative sample pair recognition mechanism, the problem of representation collapse can be alleviated to a certain extent. However, complex negative sample generation operations will Occupies memory space and affects GPU running speed

In this case, the insufficient discriminative ability of node representation will lead to limited clustering performance, which limits the performance of existing GCN-based clustering algorithms.

Images