Chinese Text Error Correction Method Based on Prefix Tree Merging

摘要

In order to solve the problem of high computational complexity and repetitive calculation of the Long Short-Term Memory (LSTM) language model in the task of Chinese text automatic proofreading, a Chinese text error correction method based on prefix tree merging is proposed in this paper. The method has made the following improvements: different from the traditional error correction method based on N-gram model, the method use LSTM language model to evaluate the rationality of the candidate sentences, and the candidate sentences with higher similarity are combined into a tree structure and then scored. Repetitive calculations can be reduced by merging the same calculations when the model calculates the probability of candidate sentences, which thereby could improve the calculation efficiency of the LSTM language model. The experimental results show that this method can not only achieve good error correction accuracy, but also shorten the time consumed and improve the error correction efficiency.