Automatic abstracting method and device based on XLNet

Abstract

The invention discloses an automatic abstracting method and device based on XLNet. The automatic abstracting method comprises the following steps: firstly, carrying out data preprocessing, segmentinga text according to sentences, segmenting sentences according to words, and adding displayed placeholders for defining sentence boundaries in front of each sentence; then, constructing an abstract model XLNetSum, wherein global position codes are added to the model on the basis of XLNet, and dynamic word vectors corresponding to placeholders and the global position codes serve as sentence features; and after the model is trained by using the training data and the verification data, obtaining scores of all sentences in the test data through the trained model, and selecting a plurality of sentences as abstracts through a post-processing step. The abstract model extracts text information through the deep neural language model XLNet, mines semantics of words and context syntax structures, processes text sequences with uncertain lengths, can flexibly and accurately judge the importance of all sentences in the text, and then extracts abstract sentences of the text.

Description

technical field [0001] The invention belongs to the technical field of information processing, and in particular relates to an automatic summarization method and device based on an XLNet model. It mainly uses the neural language model XLNet to extract the semantic information of the text, which overcomes the problem of insufficient extraction of word semantics and syntactic structure information by traditional methods, and does not limit the sequence length of the input text, and can flexibly and accurately extract summary sentences from the text. Background technique [0002] With the continuous development of Internet technology, especially mobile Internet technology, people's study, work, life and other aspects are closely related to the Internet. Information on the Internet has brought convenience to people, but in the face of a huge amount of network information, it is difficult for people to select the most useful information for themselves. Automatic text summarizatio...

AI Technical Summary

Problems solved by technology

Among them, XLNet adopts the relative position encoding of words calculated, which theoretically supports infinitely long document sequence modeling, but it does not consider the position information of sentences in documents, which may have an adverse impact on document summarization tasks.

Images