Applying Deep Learning Approach to Targeted Aspect-based Sentiment Analysis for Restaurant Domain

Abstract

Sentiment analysis is a well-studied topic in social media analysis and it becomes an important decision-making tool to classify the opinion of user on the Web. In a few years ago, there are so many research works for sentiment analysis using machine learning. Nowadays, deep learning models are implemented in AI applications to gain better performance. This paper proposes to extend the traditional LSTM approach by adding the external knowledge. LSTM is a kind of recurrent neural network (RNN) and it can learn the past and future information. Unlike traditional RNN, it can capture the long sequence of text. With the LSTM cell, SenticNet is used as an external knowledge, to improve the accuracy in classification. We termed SenticNet multi-attentive LSTM (MA-LSTM). The results from the SenticNet MA-LSTM are classified as positive, negative or neutral and finally, compare the results with other state-of-the-art LSTM methods: standard LSTM, TD-LSTM, TC-LSTM, AE-LSTM, and ATAE-LSTM. To our knowledge, this is the first work for using multi-attention LSTM with external knowledge in targeted aspect-based sentiment analysis task.