Group discovery method based on depth map convolutional network

Abstract

The invention discloses a group discovery method based on a depth map convolutional network. The group discovery method is used for solving the problem that an existing method is low in group structure recognition rate in an attribute network. The method specifically comprises the following steps: acquiring attribute network user interaction behavior data; modeling an attribute network topology and determining a small number of node tags by preprocessing the attribute network data; pre-training existing node tags by using a tag propagation algorithm to expand a tag set; and constructing a depth map convolution model, performing deep fusion on structure information and node attributes at the same time, and automatically identifying a complete group structure. By adopting the technical scheme of the invention, mining of group features in a large-scale attribute network is facilitated, and meanwhile, the accuracy of group identification is effectively improved.

Description

technical field [0001] The invention belongs to the field of network data mining. Specifically, it involves a method for group discovery based on deep graph convolutional networks. Background technique [0002] In recent years, with the rapid development of information technology and the Internet, the connection and interaction between people and between people and the environment have become common and complex, thus forming a variety of complex systems. These complex systems can usually be described abstractly by complex networks, such as online social networks, mobile communication networks, etc. Complex networks involve many interdisciplinary fields such as physics, biology, social science, system science, and network science, and have gradually become a powerful tool for solving complex problems. Data analysis and many other fields have a wide range of applications. The network topology formed by interconnected individuals in these complex network systems is random an...

AI Technical Summary

Problems solved by technology

However, such methods are deficient in terms of model structure and scalability

On the one hand, such methods usually identify unknown nodes in a supervised or semi-supervised manner, which requires a large number of node labels for model optimization during model training, and it is difficult to obtain a large number of prior labeled nodes in practical applications, or There are only a small number of labeled nodes

On the other hand, due to the problem of over-smoothing caused by graph convolution, the currently used graph convolutional network model actually contains a shallow structure of 2 to 3 convolutional layers, which cannot take advantage of the global network in the node label inference process. information

Images