Research And Application Of Chinese Text Error Correction Method

Text error correction is one of the important applications of natural language processing technology.The two-stage method of error detection and correction is a commonly used method for Chinese text error correction.The method firstly detects the error position and type in the text,and then corrects the error according to the error detection result.Currently,the performance of Chinese text error detection methods is low and the training process is complex;Chinese text error correction methods usually corrects misspelled words only,and ignores missing errors and word order errors correction.For Chinese text error detection,this paper proposes an error detection method that fuses adjacent character features,and builds a sequence labeling error detection model based on ELECTRA and Gated-Bilinear-GNN.On the basis,the graph neural network can effectively utilize the characteristics of the relative position information of nodes,strengthen the local semantic correlation of word vectors,and improve the performance of error detection.Experiments show that the performance of this error detection method reaches the state-of-the-art in both single-model and model ensemble methods.For Chinese text error correction,this paper uses a variety of strategies to deal with different kinds of errors.When correcting misspelled errors,this paper uses the BERT masked language model and beam search under the constraints of phonetic and glyph confusion sets.And the performance of misspelled word correction is greatly improved;when correcting missing and word order errors,this paper uses the BART sequence generation model to fill in blanks or reorder words,and use special tags to constrain the sequence generation results,alleviating the hallucinations of the Seq2 Seq model.Experiments show that the performance of multi-strategy error correction methods for different types of errors is higher than that of the comparison models.Finally,based on the above Chinese text error detection and correction methods,this paper conducts research and analysis on user needs,designs and implements a Chinese text error correction prototype tool,and provides a complete set of solutions for the application of Chinese text error correction.