Insufficient redundancy or excessive redundancy of information in parallel corpora can interfere with the semantic relevance analysis in the process of machine translation and affect the quality of machine translation. For this reason, a neural machine translation model has been designed that incorporates information entropy to determine base weights and SVM to eliminate redundant samples for scientific and technical English texts. The results of simulation experiments show that the present method (RNNSearch +IE+ SVM) improves the BLEU value by 1.06 BLEUs compared to the baseline model in the English-German translation task. SVM excels at parsing both under-redundancy and over-redundancy in binary classification experiments for similarity. The research has the potential to significantly enhance the efficiency of neural machine translation for scientific and technical English texts and achieve excellent experimental results, paving the way for new ideas and methods for neural machine translation research.