Neural machine translation method based on Transform model optimization

The invention provides a neural machine translation method based on Transform model optimization, and the method comprises the steps: building a translation model, obtaining a Chinese-English data set in News Common v12 in WMT 17, and carrying out the normalization processing of the data; sentence structure information is introduced into the translation model, word vectors are obtained, and a new input sequence is obtained; training a translation model by utilizing the sequence fusing the position information and the syntactic information, and performing optimization iteration by adopting an adam optimizer to obtain a neural machine translation model; inputting the test set into the neural machine translation model optimized in the step S3 to obtain a translation result; evaluating a translation result, and ending translation; and using the BLEU value as a judgment criterion of the translation model. According to the method, the translation accuracy of Chinese-English neural machine translation can be improve