Deep Learning in Aspect-Based Sentiment Analysis

Abstract

Sentiment analysis helps to capture the community’s opinion about a trend, product, service or public figure. The purpose of sentiment analysis is to build a system of extracting, classifying and defining opinions in reviews. Aspect-based sentiment analysis pays attention to the sentiment extract corresponding to the aspect of the entity in the sentiment text. In this paper, we propose a novel supervised learning approach using deep learning techniques with GRU, and CNN for aspect-based opinion mining system. Our approach focuses on extracting aspects and classifying respective opinions in review. We use GloVe with 100-dimension feature vector for word embeddings and experiment our proposed model on Restaurant and Laptop domains of the SemEval 2014 benchmark dataset. Experimental results showed that our model has extracted and classified simutaneously aspect and opinion in sentiment text and achieved better accuracy than the previous state-of-the-art models.