CE-HEAT: An Aspect-Level Sentiment Classification Approach With Collaborative Extraction Hierarchical Attention Network

Aspect-level sentiment classification is a fine-grained sentiment analysis task. It aims to predict the sentiment of a review in different aspects. Recently many works exploit the hierarchical attention network to capture the aspect-specific sentiment information. However, the prior work only attends to use the aspect terms to capture the aspect-specific sentiment information in the text. It may cause the mismatch of sentiment for the aspect-specific when the aspect words are extracted incorrectly. Since the number of aspect words is much more than that sentiment words, some uncommon aspect terms are more difficult to be extracted than uncommon sentiment terms. To solve this problem, we propose a collaborative extraction hierarchical attention network. It consists of two hierarchical attention units, i.e.. One unit extracts the sentiment features by the sentiment attention layer and uses the sentiment features to capture the specific aspect-related sentiment aspect information by the aspect attention layer. The other extracts the aspect features by the aspect attention layer and uses the aspect features to help capture the aspect-specific sentiment information by the sentiment attention layer. Moreover, we use the SemEval competition data set to verify our approach. The experiment shows that our approach achieves better performance than the methods which only use aspect features to extract sentiment feature for aspect-level sentiment classification.