Aspect-Based Sentiment Analysis of Indonesian-Language Hotel Reviews Using Long Short-Term Memory with an Attention Mechanism

Abstract

The development of tourism and technology has given rise to many online hotel booking services that allow users to leave reviews on hotels. Therefore, an analytical model that can comprehensively present the aspects and sentiments in user reviews is required. This study proposes the use of a long short-term memory (LSTM) model with an attention mechanism to perform aspect-based sentiment analysis. The architecture used also implements double fully-connected layers to improve performance. The architecture is used simultaneously for aspect extraction and sentiment polarity detection. Using 5200 Indonesian-language hotel-review data points with labels of five aspects and three sentiments, the model is trained with the configuration of hidden units, dropout, and recurrent dropout parameters in the LSTM layer. The best model performance resulted in a micro-averaged F1-measure value of 0.7628 using a hidden units parameter of 128, dropout parameter of 0.3, and recurrent dropout parameter of 0.3. Results show that the attention mechanism can improve the performance of the LSTM model in performing aspect-based sentiment analysis.