Structurally Enhanced Interactive Attention Network for Aspect-Level Sentiment Classification

Aspect sentiment analysis main purpose is to identify it in the context of the target of sentiment polarity. Previous approaches have recognized the importance of syntactic dependency tree structure information for sentiment classification. However, the target word associated with the sentiment expression of nodes for the target words is particularly important, while the rest of the edge nodes played an auxiliary role. Therefore, we intend to extract the side nodes associated with the target and then interactively express them with the context. Through this kind of structurally enhanced expression, the context and dependence information of the two are transmitted from opinion words to aspect words, which strengthens the discriminability of supervision. The experimental results on the 2014 datasets verify the effect of our model.