CROSS-LANGUAGE SENTIMENT CLASSIFICATION METHOD BASED ON ADVERSARIAL BIDIRECTIONAL GATE RECURRENT UNIT NETWORK

In order to improve sentiment classification performance of resource-scarce languages, a cross-lingual sentiment analysis classification model (ABi-GRU) based on the combination of adversarial bidirectional GRU network is proposed. The model extracted the word vector features of Chinese and English texts based on semantic bilingual word embedding. Combining with the bidirectional GRU network of attention mechanism, the text’s contextual emotion features were extracted and the generative adversarial network was introduced to narrow the gap between Chinese and English vector feature distribution. The sentiment classification was carried out by sentiment classifier. Experimental results show that this method effectively improves the accuracy of cross-language sentiment classification.