A Transfer Learning based Approach for Aspect Based Sentiment Analysis
Xing Fang, Jie Tao · 2019
Understanding sentiments embeded in online consumer reviews can assist a variety of decision making processes. Recent studies have extended sentiment analysis to a fine-grained level, such as toward different features or characteristics of products/services, which is known as Aspect Based Sentiment Analysis (ABSA). In this study, we propose a transfer learning based multi-label classification approach, building on a pre-trained advanced deep learning model. We evaluate the proposed approach in an experiment by comparing it with mainstream multi-label classification approaches in the context of ABSA on Yelp restaurant reviews. To the best of our knowledge, this is the first study conducting multi-label classification of ABSA on online consumer reviews. The experiment results confirm that our approach outperforms other multi-classification approaches.