Neural Attentive Network for Cross-Domain Aspect-level Sentiment Classification

摘要: This work takes the lead to study aspect-level sentiment classification in domain adaptation scenario. Given a document of any domains, model needs figure out sentiments with respect fine-grained aspects documents. Two main challenges exist this problem. One is build robust modeling across domains; other mine domain-specific and make use lexicon. In paper, we propose novel approach NAACL (Neural Attentive for cross-domain Aspect-level CLassification), which leverages benefits supervised deep neural network as well unsupervised probabilistic generative strengthen representation learning. evaluated on both English Chinese datasets out-of-domain in-domain setups. Quantitatively, experiments demonstrate that has superiority over compared methods terms accuracy F1 score. The qualitative evaluation also shows proposed capable reasonably paying attention those words are important judge polarity input text given an aspect.