A Model Of Pinyin Input Method With Error Correction Function Based On Neural Network

With the rapid popularization of mobile devices,people are increasingly inseparable from intelligent devices to communicate and exchange information.Pinyin input method is an important tool in daily life.However,in the use of the input method will inevitably produce wrong spelling.The spelling error correction problem of pinyin input method greatly affects the user experience.This paper firstly analyzes the key technologies used in this project.With the concern of the structural characteristics of Chinese pinyin spelling error correction problem in the input method,this paper proposed a neural network Chinese pinyin typo correction model with the combination of external information based on Seq2seq model.The loss function of seq2seq model is improved according to the the key transfer probability of the keys on the keyboard and the alignment relationship between the input and output of the input method,which realizes the supervised training of the Attention mechanism in the error correction model and improve the error correction ability of the model.In addition,aiming at the abundant priori knowledge in pinyin input method,a method combining priori knowledge with neural network is proposed to optimize the error correction model.The vector expression of click position through Autoencoder method,improving the input layer of neural network.This can make the pinyin typo correction model effective use the click location information,and thus has stronger error correction ability for the spelling error input caused by user's wrong touch.In addition,as the Beam Search algorithm is used in the decoder part of the pinyin spelling error correction model,thus the model output is a number of pinyin candidate set.In view of the unreasonable ordering of pinyin candidates,a statistical characteristics of pinyin language based reordering model was proposed.A improved pairwise algorithm is used to score the pinyin candidates,and the high-quality candidate of pinyin was ranked at the top of the candidate set,further improve the accuracy of error correction results.Finally,a system of pinyin input method with error correction function is realized by combining the above model and algorithm.The system includes a pinyin spelling error correction module and a Chinese character to pinyin conversion module based on hidden markov model.Finally,the whole system is constructed and verified.The results show that the proposed model of pinyin input method with error correction function based on neural network has complete functions and strong spelling error correction ability.The optimization and improvement of the spelling error corrention model in this project can effectively improve the user experience of the input method and has practical application value.