Semi-supervised Target-level Sentiment Analysis via Variational Autoencoder.

摘要: Aspect-term sentiment analysis (ATSA) is a longstanding challenge in natural language understanding. It requires fine-grained semantical reasoning about target entity appeared the text. As manual annotation over aspects laborious and time-consuming, amount of labeled data limited for supervised learning. This paper proposes semi-supervised method ATSA problem by using Variational Autoencoder based on Transformer (VAET), which models latent distribution via variational inference. By disentangling representation into aspect-specific lexical context, our induces underlying prediction unlabeled data, then benefits classifier. Our classifier agnostic, i.e., an independent module various advanced can be integrated. Experimental results are obtained SemEval 2014 task 4 show that effective with four classical classifiers. The proposed outperforms two general semisupervised methods achieves state-of-the-art performance.