Interactive Attention and Position-Aware Mechanism for Aspect-Level Sentiment Analysis

Aspect-level sentiment classification is a fine-grained work in sentiment analysis with the goal of predicting the sentiment categories of target words in a given text. In this paper, a framework of an interactive attention based on bidirectional LSTM networks and position-aware mechanism for aspect-level sentiment classification has been proposed, in which a fine-grained attention is introduced to capture the intersection between context and aspect words. Meanwhile, the position awareness mechanism is introduced to make the weight distribution of cross attention more reasonable. Experiments on the SemEval-2014 datasets show that our proposed model is always superior to the state-of-the-art methods of all datasets.