OPTIMIZATION OF MACHINE TRANSLATION MODEL BASED ON DBOA-BP NEURAL NETWORK

To enhance the translation quality of neural machine translation (NMT), a developed butterfly optimization algorithm (DBOA) and back propagation (BP) neural network were applied to optimize the dropout and Learning_data parameters in the machine translation model. Then, a neural machine translation model was built based on sequence-to-sequence (Seq2Seq) model with attention mechanism, and the training data of BP neural network was obtained after many times of training. Meanwhile, the key parameters of the neural network translation model were optimized by DBOA, which was mainly improved by two strategies: changing weights dynamically and adjusting switch coefficients of searching mode dynamically. With the Bleu value as the fitness value, DBOA was combined with the BP neural network to optimize the dropout and Learning_data parameters in the NMT model to achieve the theoretical optimal Bleu value. The dropout values and Learning_data solved by the algorithm were substituted into the NMT model to get the true Bleu value, which was approximately the same as the predicted value. Thus, the parameters of dropout and Learning_data in the neural translation model were effectively optimized, so the translation quality was developed to a certain extent.