Text reconstruction training method and system

Abstract

The invention discloses a text reconstruction training method and system. The method comprises the following steps: constructing a training sample; encoding splicing of the original text and the pronunciation sequence by adopting a neural network text sequence model; adopting a sigmoid feedforward neural network classifier to obtain a word segmentation classification result; taking a preset GEN _LENGTH as the maximum generation sequence length, and circularly generating each character of the sequence from head to tail; optimizing parameters of the neural network, calculating word segmentationloss and text generation loss, performing weighted summation on the generation loss to obtain joint loss, and optimizing the parameters of the neural network for the joint loss by using a gradient descent algorithm. According to the text reconstruction training method and system provided by the invention, the text coding and the generative neural network are used for converting the given originaltext into the target text, correcting wrongly written characters in the original text, supplementing missing characters, removing redundant characters and standardizing words, so that the purposes ofeliminating text errors and improving the text quality are achieved.

Description

technical field [0001] The invention relates to the technical field of text generation and deep learning, in particular to a text reconstruction training method and system. Background technique [0002] In the information age, text is an important information carrier in Internet multimedia, with huge data and many sources and authors. Due to problems such as similar characters in shape, similar in sound, clerical errors, slips of the tongue, inaccurate speech recognition, and uneven levels of authors, the phenomenon of wrong words and irregular words in the text occurs from time to time, causing misinformation and misunderstandings . The existing text error correction methods based on tree models or deep neural networks cannot effectively solve the problems of missing words, multiple words, mispronunciation of speech recognition results, and irregular words in the text. Contents of the invention [0003] The purpose of the present invention is to provide a text reconstru...

AI Technical Summary

Problems solved by technology

Due to problems such as similar characters in shape, similar in sound, clerical errors, slips of the tongue, inaccurate speech recognition, and uneven levels of authors, the phenomenon of wrong words and irregular words in the text occurs from time to time, causing misinformation and misunderstandings

The existing text error correction methods based on tree models or deep neural networks cannot effectively solve the problems of missing words, multiple words, mispronunciation of speech recognition results, and irregular words in the text.

Images