Research on the application method of NLP technology based on AutoPrompt

Sentiment analysis methods usually rely on a large amount of labeled data, but in practice, due to the limitation of corpus resources, most of the existing sentiment analysis methods are based on English texts, and there are few studies on languages with scarce corpus resources. On the other hand, for the same task in different languages, each task has to do feature engineering again. The research on language-specific feature and representation extraction is continuously refined, and re-extracting features is very time-consuming and labor-intensive. The purpose of this paper is to find a way to learn common expressions in different languages, so that sentiment analysis of unlabeled text in target language can be done using only labeled data in the source language. The paper is based on the task of sentiment analysis, through adversarial training on the basis of the pre-training model, and at the same time in order to alleviate overfitting and catastrophic forgetting caused by the lack of target language labels, a knowledge distillation method is used for joint training. In order to further strengthen the “sentiment” factor of the model, “emoji” is used as an auxiliary label to learn the unique expression patterns of different languages, and further improve the effect of cross-language sentiment analysis.