Opinion Knowledge Injection Network for Aspect Extraction

Aspect term extraction (ATE) is to extract explicit aspect expressions from online reviews. This paper focused on the supervised extraction of aspect term. Previous models for ATE either ignored the opinion information or improperly utilized the opinion information with a high-coupling method. We proposed a model to perform ATE with the assistance of opinion knowledge, called opinion knowledge injection network. Specifically, the proposed model distills the opinion knowledge through the attention mechanism and joins it into each word to assist aspect extraction. The proposed model achieved surprisingly good results, improving 1.34% and 1.23% than the best results before respectively on the laptop and restaurant datasets, and reached state-of-the-art.