Research On Key Techniques Of Chinese Grammar Error Correction Based On Neural Network

Computer-assisted learning is a kind of computer technology that is often used by teachers to use computers to help self-learning,break through difficulties,and practice repeatedly during the teaching process.At present,the main research hot spots of Chinese assisted learning technology mainly focus on two aspects.One is the correction of Chinese grammar and the other is the correction of Chinese typos.However,the current model faces many problems such as poor generalization ability,a large amount of manual annotation,fewer words and phrases in the local word bank,and a more complicated model.For the appeal issue,this article is based on gated recursive unit and conditional random field,as well as the convenience of the Internet,to study the characteristics of integrating words,words,and part of speech in Chinese,analyzing the composition of sentences,and obtaining more words from the Internet.The main contents include Two aspects:(1)In order to determine Chinese grammatical errors,this thesis proposes a combined model based on gated recursive unit and conditional random field(GRU_CRF).This model uses the character vector to represent sentence features,integrates sentence features through GRU,classifies grammatical errors,and finally use CRF to determine where the error occurred.It solves the problem of ambiguity caused by word segmentation in Chinese sentences,poor fitting of text features,and the large amount of manual annotation required for using CRF models alone.The comparison experimental results show that the model is superior to the existing machine learning model and other neural network models.(2)For the correction of typos,this thesis uses a part-of-speech tag to use the CRF model to fit Chinese sentence collocation features.It solves the problem that the existing model needs to formulate a large number of rules and has poor generalization ability.At the same time,the network word database is introduced in the experiment,which improves the problem that many words can't be matched due to too few words in the local word bank.The experimental results show that a simple method achieves better results.