Hypergraph-based text classification method

Abstract

The invention discloses a hypergraph-based text classification method, which comprises the following steps of: 1, constructing a corpus and a keyword library of the corpus, and generating a hypergraphbased on the corpus to obtain hyperedges and nodes of the hypergraph; 2, calculating an adjacency matrix of the keywords based on a co-occurrence window; 3, expressing each document in the corpus through a hyperedge vector to form a hyperedge matrix; 4, calculating the similarity between the hyperedges, and constructing a similarity matrix of the hyperedges; step 5, constructing a hypergraph nodefeature matrix composed of word vectors; 6, classifying the hyperedges by using a graph neural network model to obtain a first prediction probability of each document category in the corpus; 7, updating a parameter matrix of the graph neural network model by adopting a stochastic gradient descent algorithm based on the real label of the document, and finishing classification of the label-free text in the corpus. According to the method, the label-free texts in the corpus are accurately classified.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a hypergraph-based text classification method. Background technique [0002] Graphs representing objects and their relationships exist everywhere in real life, such as social networks, e-commerce networks, biological networks, and transportation networks. At the same time, due to the rich potential information contained, the graph is also recognized as a structure that can be deeply understood. In the past decade, graph deep learning has become an extremely important part of artificial intelligence and machine learning. It has shown superior performance in audio, image and natural language processing, and has obvious effects in extracting potential complex patterns in data. [0003] The essential feature of a hypergraph lies in its hyperedges. The degree of a hyperedge can be greater than 2, and it can connect multiple nodes at the same time. Ordinary graphs are a speci...

AI Technical Summary

Problems solved by technology

[0004] Existing text classification methods usually regard the texts in the corpus as independent samples, use models such as recurrent neural networks and support vector machines to predict the category for each piece of text independently, and do not model the relationship between texts

In addition, when using a cyclic neural network for text classification, when the text length is too long, it is difficult for the cyclic neural network to capture the meaning expressed by the text, and the model effect is poor

Images