Sentiment analysis and validity evaluation of Japanese language under the transfer learning model

Sentiment analysis of language is an important part of natural language processing, which is of practical significance for promoting human-computer interaction in the era of artificial intelligence. In this paper, we improve classical sentiment analysis and transfer learning methods and propose a multi-step migration method based on RoBERTa. After preprocessing the Japanese text, the cross-entropy loss function is mapped and the gradient descent rate is adjusted in the training of the RoBERTa model to solve the negative migration and catastrophic forgetting problems, and the RoBERTa is fine-tuned after the training is completed. Compared with the non-migration methods, the accuracy of Japanese sentiment recognition using the migration learning method is higher, and the multi-step migration learning method based on RoBERTa in this paper in turn has better results for Japanese sentiment analysis than the general migration method (accuracy = 83.71%, F1 = 74.5%). Thus, the effectiveness of this paper’s method for sentiment analysis of Japanese language is fully realized.