Cross-lingual Sentiment Analysis via AAE and BiGRU

As a significant task in nature language processing (NLP), sentiment analysis has got more and more attention. Sufficient data and helpful tools are available when doing the studying. Therefore, it is meaningful to study techniques to leverage labeled data and models valid in rich datasets language when addressing any questions in rare language. As a result, cross-lingual sentiment analysis has become more and more popular. Compared with traditional translation method, transfer learning is the main trend and using deep learning to generate a cross-lingual word embeddings in the single vector space is more useful and stable. Hence getting the high quality of cross-lingual word embedding is the big problem which needs to be settled urgently. In this paper, we use the LSTM and Adversarial Auto Encoders (AAE) to generate contextual cross-lingual word embeddings for transfer learning, then the BiGRU is used for analyzing the sentiment. Experimental results prove that our method has the best performance.