Short text topic discovery method and system based on self-expanding representation and similar bidirectional constraints

Abstract

This invention clearly provides a short text topic discovery method and system based on self-expanding representation and similar two-way constraints, the steps of the method include: constructing a word-document matrix of the topic to be mined based on the TF-IWF word weight measurement method; vectorizing the short text document and measure the similarity between two documents to obtain a virtual long document collection; use the TF-IWF word weight measurement method to construct a virtual auxiliary word-document matrix on the virtual long document collection; merge the two matrices into a mixed matrix; construct a word -word semantic similarity matrix, document-document semantic similarity matrix, and then construct word-word semantic relationship regularization items, document-document semantic relationship regularization items; get the TRNMF model, and obtain the optimal word-topic latent features by decomposing the loss function value Matrix, topic-document latent feature matrix, discovers the distribution of short text topics.

Description

technical field [0001] The present invention relates to a topic mining technology of social short text data which contains a social network, in particular to a short text topic discovery method and system based on self-expanding representation and similar bidirectional constraints. Background technique [0002] With the rapid development of the Internet, mobile applications and social networks, massive short text messages have shown explosive growth. Analyzing and mining the deep semantic structure behind these unstructured text information has important theoretical value and practical significance. For example, accurate semantic understanding can help enterprises improve product functions and user experience according to users' search request preferences; help the government detect harmful information and prevent crises, which plays an important role in stabilizing society; help users avoid information overload problems, filter Useless information, only focus on valuable i...

AI Technical Summary

Problems solved by technology

From the perspective of the co-occurrence relationship of words inherent in the data, the TRNMF model uses the regularized non-negative matrix factorization algorithm as a basis for self-expanding representation of short texts, which solves the problem of data sparsity, thereby improving the topic modeling of short texts precision and efficiency

Images