Double-flow graph convolutional network microblog topic detection method fusing different propagation modes

Abstract

The invention discloses a double-flow graph convolutional network microblog topic detection method fusing different propagation modes. The method comprises the following steps: (1) constructing a user-level social network according to a user interaction relationship; (2) for different propagation modes, aggregating attribute information of related nodes of each user node by using a message passing mechanism of a graph convolutional network, wherein learning user node embedding representation contains specific propagation mode characteristics; and (3) splicing the user node embedded representations containing the two propagation mode characteristics, generating potential topic vectors and topic distribution by using an encoder part in the variational auto-encoder, training topic-word distribution by using a decoder part, and reconstructing the user node embedded representations. According to the method, more complete social context information is modeled, better user node embedding representation is learned, and more coherent topics are generated. The experimental result is better than that of the existing model.

Description

technical field [0001] The invention relates to the technical fields of natural language processing and social media data mining, and specifically relates to a method for detecting topics of microblogs in a dual-stream graph convolutional network that integrates different propagation modes. Background technique [0002] With the popularity of social media such as Twitter and Sina Weibo, countless short texts are generated on the Internet every day. These texts contain a wealth of information such as user opinions and viewpoints. Analyzing the content of these posts manually is a daunting task, time-consuming and labor-intensive. Topic models are a common tool for automatically analyzing massive texts. It can automatically detect topics from documents and output document-topic distribution and topic-word distribution. Traditional topic models infer topics based on rich word co-occurrence patterns in documents. They use Markov chain Monte Carlo (MCMC) or Expectation-Maximu...

AI Technical Summary

Problems solved by technology

[0005] Although the above methods have achieved good results, they ignore the mining of communication features when modeling social context

Images