Text error correction method and device

Abstract

The present application proposes a text error correction method and device, wherein the method includes: input the word vector array corresponding to the text to be error corrected into the preset encoding module, obtain the first hidden state vector array and input it to the decoding module, and for each The decoding position determines the decoding vector according to the second hidden state vector, the attention vector and the first hidden state vector array corresponding to the decoding position; The limited word table determines the decoding result of the decoding position, and then determines the error-corrected text corresponding to the text. When determining the decoding vector in this method, the first hidden state vector array is used, thus taking into account the word order of the text, ensuring corrected text. In addition, the use of restricted vocabulary limits the size of the understanding space, reduces the complexity of the error correction model, and improves the convergence speed of the model.

Description

technical field [0001] The present application relates to the technical field of data processing, in particular to a text error correction method and device. Background technique [0002] The current end-to-end error correction model is the NMT error correction model based on the introduction of attention mechanism. The NMT error correction model is a Sequence-To-Sequence model based on Encoder-Decoder. Among them, the structure of the Encoder module and the Decoder module is a recurrent neural network (Recurrent Neural Network, RNN for short) network structure, and the vocabulary used by the two to map the words / segments in the text sequence to the word vector space is the same. However, since the magnitude of the vocabulary is usually tens of thousands to hundreds of thousands, the solution space of the error correction model is too large when decoding the output, the complexity of the model is high, and the convergence speed of the model training is too slow. Moreover, ...

AI Technical Summary

Problems solved by technology

However, since the magnitude of the vocabulary is usually tens of thousands to hundreds of thousands, the solution space of the error correction model is too large when decoding the output, the complexity of the model is high, and the convergence speed of the model training is too slow.

And the attention mechanism does not consider the order of the original input sequence of the Encoder module, resulting in poor accuracy of error correction results

Images