Research on Japanese Translation Quality Evaluation Model Based on Big Data

The Japanese Translation Quality Evaluation Model (JTQEM), powered by big data, combines bidirectional LSTM and attention mechanism to address the growing demand for automated translation quality assessment. Based on 53,721 Japanese-Chinese parallel corpora and MeCab morphological analysis, it builds a comprehensive evaluation dataset across 8 domains including scientific, medical, and literary texts. The model achieves $91.2 \%$ consistency with human evaluation, processes 37.2 samples per second, and reaches $94.2 \%$ accuracy in scientific text evaluation. Improvements like hierarchical attention mechanism enhance the model’s long-sentence processing capability and training efficiency. The system implements domainadaptive feature extraction, achieving significant performance gains across different text types. Advanced optimization strategies, including dynamic batch normalization and label smoothing, further improve model convergence and reduce resource consumption while maintaining high accuracy in real-world applications.