Method for generating guided text abstract based on Transformer

Abstract

The invention relates to a method for generating a guided text abstract based on Transformer, and belongs to the technical field of information processing. According to the invention, a deep learningalgorithm and a machine learning algorithm are combined to solve the problem of automatically obtaining the text abstract under the big data condition. Firstly, a text key semantic feature extractionmethod is constructed, and key semantic features of a text are obtained through the method; the method also includes, secondly, converting the long text into a key short text in combination with an extraction type abstract method to serve as input of an abstract model; and finally, constructing a text abstract generation model based on Transformer by utilizing the extracted text key semantic features. In the abstract generation model, the attention mechanism is corrected by using the key semantic features, so that the generation model can generate more abstract contents rich in key information, and a pointer and a coverage mechanism are added, so that the abstract generation model can better solve the OOV problem and the repeated fragment generation problem in the abstract generation process.

Description

technical field [0001] The invention belongs to the technical field of information processing, and relates to a Transformer-based guiding text summary generation method. Background technique [0002] Automatic text summarization is the process of extracting the most critical information from the original text, and then constructing the important content required by users. The automatic text summarization method refers to the use of machines to automatically summarize a concise and readable summary sequence rich in key information from a large amount of text data. At present, from the perspective of construction methods, automatic text summarization mainly includes two types: generative summarization and extractive summarization. Extractive summarization uses some statistical methods to sort the importance of all sentences in the original text, and then extracts several important sentences with the highest importance as the summary of the text. Generative summarization refe...

AI Technical Summary

Problems solved by technology

[0004] The traditional Seq2Seq (Sequence to Sequence) framework based on Recurrent Neural Networks (RNN) can only input one word at a time, has low computational efficiency, and is less dependent on long texts

Images