Automatic text abstracting method based on heterogeneous graph network

Abstract

The invention relates to the technical field of text processing, in particular to an automatic text abstracting method based on a heterogeneous graph network. The method includes the following steps: S1, representing an initial word vector of each word by using pre-trained Word2Vec; S2, capturing word context semantic information by using CNN and a window with the length of n to finally obtain a vector representation of a jth sentence, and meanwhile, performing sequence modeling on a word sequence in each sentence by using LSTM to finally obtain a vector representation of each sentence; S3, iteratively updating the heterogeneous graph: updating the heterogeneous graph network by using a graph attention mechanism; and S4, scoring and sorting the vector table of the sentences obtained through the heterogeneous graph, and selecting sentences suitable for being used as abstracts. Through the steps, each sentence in the original text can obtain an important score, proper abstract sentences can be extracted according to score sorting, and the method is mainly used for automatic abstracting of texts.

Description

technical field [0001] The present invention relates to the technical field of text processing, and more specifically, to an automatic text summarization method based on a heterogeneous graph network. Background technique [0002] Existing techniques for text extraction and summarization all use sequence modeling to model text. This method will ignore the deep-level relationship between many texts, resulting in poor extraction effect. In fact, words between different sentences also have a corresponding relationship, but sequence modeling cannot accurately represent long-distance information well, so it cannot associate the same information between two different sentences, resulting in semantic deviation. The current extractive summarization methods are all based on the LSTM modeling method, which models the text from left to right. This method can learn the language model well, but it cannot effectively capture long-distance text interaction information. Therefore, it is ne...

AI Technical Summary

Problems solved by technology

[0002] Existing techniques for text extraction and summarization all use sequence modeling to model text. This method will ignore the deep relationship between many texts, resulting in poor extraction effect.

In fact, words between different sentences also have a corresponding relationship, but sequence modeling cannot accurately represent long-distance information well, so it is impossible to associate the same information between two different sentences, resulting in semantic deviation

The current extractive summarization methods are all based on the LSTM modeling method, which models the text from left to right. This method can learn the language model well, but it cannot effectively capture the long-distance text interaction information.

Images