SAHAR-LSTM: An enhanced Model for Sentiment Analysis of Hotels'Arabic Reviews based on LSTM

Over the last few years, many scientists have paid special attention to the Sentiment analysis (SA) area of research, thanks to its interesting uses in different domains. The most popular studies have tackled the issue of SA in the English language; however, those dealing with SA in the Arabic language are, up to now, limited due to the complexity of the computational processing of this Morphologically Rich Language (MRL). As a matter of fact, Deep learning and especially the use of Recurrent Neural networks (RNN) has recently proved to be an efficient tool for handling SA challenges. The recourse to some approaches, based on the long short-term memory (LSTM) architecture, has provided adequate solutions to the problems of SA in Arabic language. In our paper, we conduct a study on SA in the Arabic language. Therefore, we propose an enhanced LSTM based model for performing SA of Hotels' Arabic reviews, called SAHAR-LSTM. This model is evaluated on a dataset containing Hotels' reviews written in Modern Standard Arabic (MSA), and it is implemented together with two Dimensionality reduction techniques: Latent Semantic Analysis (LSA) and Chi-Square. The experimental results obtained in this work are promising, and demonstrate that our proposed approaches achieve an accuracy of 83.6% on LSA and Chi-Square methods and 92% on LSTM classification Model.