Short text similarity calculation method based on graph neural network

Abstract

The invention discloses a short text similarity calculation method based on a graph neural network. The method comprises the following steps: constructing a text graph expression to obtain basic graph structure data of a text; based on an attention-based edge learning mechanism, learning edges of the text graph through node information; adopting a mixed multi-head graph learning mechanism to perform mixed expression on nodes generated by the graph attention network and the GraphSAGE network; and calculating the graph similarity. According to the method, the obtained text representation is modeled into a graph structure, the graph structure is sent to the graph attention network and the GraphSAGE network for learning, the learned results are fused, and the final expression of the sentence integrating context information and high-level semantic information is obtained through the full connection layer. And finally, calculating the similarity between the sentence expressions by adopting a Pearson's correlation coefficient pair.

Description

technical field [0001] The invention belongs to the technical field of language processing, and in particular relates to a short text similarity calculation method based on a graph neural network. Background technique [0002] In daily life, due to the frequent appearance of short texts in the fields of Weibo, SMS, and short videos, the demand for short text similarity calculation is increasing. Short text similarity calculation is a difficult and hotspot in the field of natural language processing (NLP) and even machine learning. It is an important task in NLP. It can be regarded as a separate task or the basis of other NLP applications. At present, short text similarity calculation methods usually use external knowledge-based methods and sequence-based methods. The external knowledge-based methods use linguistic tools such as semantic association knowledge bases, semantic analysis trees, external corpora, and pre-trained models. The method relies on artificially construct...

AI Technical Summary

Problems solved by technology

At present, short text similarity calculation methods usually use external knowledge-based methods and sequence-based methods. The external knowledge-based methods use linguistic tools such as semantic association knowledge bases, semantic analysis trees, external corpora, and pre-trained models. The method relies on artificially constructed established rules and artificially constructed feature engineering, which usually requires a large amount of computing resources and professional knowledge as preliminary work; sequence-based methods usually represent a text as an ordered combination of a set of signs, which lacks Structural knowledge of the text and the relationship between two long-distance signs cannot be obtained, lacking an accurate and detailed understanding of the text content

Images