Chinese sentence simplification model training method and Chinese sentence simplification method and device

Abstract

The invention provides a Chinese sentence simplification model training method and device and a Chinese sentence simplification method and device. The Chinese sentence simplification model training method comprises the steps that a data set of complex sentence-simple sentence pairs containing supervision signals and a Chinese single-language pre-training model are obtained; selecting a simple sentence in the current complex sentence-simple sentence pair as a positive sample in each training batch, and randomly selecting a preset number of simple sentences from other sentence pairs in the same training batch as negative samples; projecting the complex sentence, the positive sample and the negative sample into a vector representation space, and respectively obtaining a hidden layer vector of the last layer of the encoder; calculating and comparing learning loss, and calculating and generating expected cross entropy loss of the simple sentences through a decoder; and the Chinese monolingual pre-training model is jointly trained by minimizing the comparative learning loss and the cross entropy loss of the simple sentences output by the Chinese monolingual pre-training model. According to the simplified model obtained by the method, the controllability and the loyalty of the generated simplified sentences can be improved.

Description

technical field [0001] The present application relates to the technical field of natural language processing, and in particular, to a training method of a Chinese sentence simplification model, a Chinese sentence simplification method, and an apparatus. Background technique [0002] Sentence reduction is to modify the content and structure of the original sentence, reduce the complexity of the sentence, and retain its main idea and close to its original semantics, so that it is easier to read and understand. Sentence reduction can be used to improve the performance of other natural language processing (NLP) tasks, such as text summarization, information extraction, and machine translation. [0003] In the related art, the early sentence reduction method adopts a rule-based method, which mainly considers lexical simplification or syntactic simplification, and the accuracy of sentence reduction is low. The current research methods are mainly divided into two types: one is to ...

AI Technical Summary

Problems solved by technology

[0006] For this reason, the first purpose of this application is to propose a training method for a Chinese sentence simplification model, which introduces contrastive learning on the basis of the pre-training model of the encoder-decoder (Encoder-Decoder) to fine-tune, by crossing On the basis of the entropy loss function, the contrastive learning loss is introduced to increase the distance between positive and negative samples, so that the trained Chinese sentence simplification model can control the generated simplified sentences according to actual needs when performing sentence simplification, and improve the generated simplified sentences. In addition, it reduces the complexity of the training process. The trained Chinese sentence simplification model can handle different Chinese sentences conveniently and efficiently. question

Images