Microblog topic detection method based on message passing and graph prior distribution

Abstract

The invention discloses a microblog topic detection method based on message passing and graph prior distribution, which comprises the following steps of: (1) on the basis of microblog corpus, constructing a user-level social network according to an interaction relationship among users; (2) performing user node embedding representation based on message passing: integrating content information and structure information of posts in social media by utilizing a graph convolutional network, and embedding an interaction relationship between users into the user node embedding representation; and (3) performing topic generation based on the graph prior variational auto-encoder: taking the user node embedded representation integrated with the user interaction relationship as input, replacing the standard Gaussian prior distribution of the variational auto-encoder with graph prior distribution containing user interaction, and considering the correlation between users in the topic inference process. In general, user interactions are integrated from two stages of user node embedding representations and topic inference. The topic detected by the method better pays attention to the correlation between the users, and higher continuity is obtained.

Description

technical field [0001] The invention relates to the technical fields of natural language processing and social media data mining, in particular to a microblog topic detection method based on message delivery and graph prior distribution. Background technique [0002] The rapid development of the Internet has brought great progress to our life. The popularity of social media allows each of us to have a platform to express our opinions and opinions. As a result, a large number of short texts are generated every day, and analyzing the topics in them is an important task, but manual analysis is time-consuming and laborious. The topic model can automatically extract document-topic distribution and topic-word distribution, helping people quickly analyze text and grasp text information. [0003] Traditional topic models, such as LDA, are widely used to discover potential topics from text corpora. Essentially, these methods reveal latent topics by implicitly capturing word co-occ...

AI Technical Summary

Problems solved by technology

[0006] While the previous methods embed user interactions into edge representations, VAEs assume that each data point is independent

Therefore, the correlation between users or posts is weakened when computing latent semantic vectors

Images