A Practical Approach Applying Deep Learning and Ontology to Identify Aspects in Opinions

Aspect-based sentiment analysis is a very interesting problem in opinion mining. Accurately determining the evaluated aspect in an opinion contributes to improving the performance of the sentiment analysis problem. This study proposes an approach to identify aspects that is not based on keywords but on the semantics of opinions. To determine the evaluated aspects in opinions, the proposed approach uses the method of embedding knowledge from the ontology into the corpus to train deep learning algorithms. The structure of the ontology used in this study is based on the relationship between aspect words and emotion words in the field of car evaluation. The corpus is labeled with aspects not only based on keywords indicating aspects but also based on the semantics of the sentence. The high accuracy test results show the prominent application of the proposed approach.