Dynamic network representation learning method and system for social network

Abstract

The invention discloses a dynamic network representation learning method and system for a social network, and belongs to the field of social network data mining. The method comprises: obtaining a Laplace matrix of the dynamic network of the social network under a plurality of continuous timestamps; for each Laplace matrix, respectively performing network representation learning by the corresponding private auto-encoder and the shared auto-encoder, splicing the two network representation results and then inputting the spliced results into the corresponding solution encoder, and restoring the Laplace matrix; constructing a loss function based on the reconstruction loss constraint, the similarity loss constraint and the difference loss constraint, and adjusting model parameters of an auto-encoder in a back propagation process by taking minimization of a loss function value as a target to complete one-time training; repeating until a termination condition is met; and taking a splicing result of the last training as a final representation result of the dynamic network under the corresponding timestamp. The high-order nonlinear structure of the dynamic network structure can be extracted,and the evolution characteristics of the dynamic social network can be accurately mined.

Description

Technical field [0001] The present invention belongs to the field of social network data mining, and more specifically, relates to a method and system for learning dynamic network representation for social networks. Background technique [0002] With the rise of the Internet, social network-related applications (such as qq, WeChat, Weibo, etc.) have become more and more popular, and data mining tasks based on social networks (such as user classification, friend recommendation, etc.) have become more and more popular. It is becoming more and more important, and dynamic network is a very important tool for representing social networks. The network data of social networks is usually complex and difficult to process, and its network structure changes in real time, and the user’s points of interest and social relations are changing smoothly. Therefore, how to mine the dynamic evolution characteristics of the historical social network structure and the low learning node The dimensiona...

AI Technical Summary

Problems solved by technology

However, the existing dynamic network representation models can only obtain the low-order linear structure of the network. However, in reality, social networks are often changing. Correspondingly, the network structure used to represent the dynamic network of the social network at different times is also It will change continuously, and only obtaining the low-order nonlinear structure of the network cannot accurately mine the evolution characteristics of the network structure, so the accuracy of downstream applications such as link prediction, node classification, and community discovery will also be affected

Images