Aspect-based Sentiment Analysis for Arabic Content in Social Media

Recently, the usage of social media platforms, especially Twitter, for sharing views and experiences toward different products and services among users has increased significantly. Accordingly, extracting and analyzing Twitter data has generated great interest from sentiment analysis researchers. Moreover, we noted the high-accuracy results achieved by the deep learning approach compared to the machine learning approach especially in analyzing a massive amounts of social networks data. In this paper, we present a survey about previous studies on sentiment analysis that used the machine learning, lexicon-based and deep learning approaches on Arabic and English tweets. Moreover, we report results based on both, machine learning and deep learning sentiment analysis approaches on Arabic tweets to extract customers' sentiments of the Saudi telecommunication companies. We also evaluate the impact of using Part of Speech (POS) model and Word Embedding on the performance of deep learning techniques. Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and Convolutional Neural Network (CNN) have been utilized to detect the sentiment orientation for a dataset consist of 1098 tweets. Results suggest that deep learning technique with Word Embedding method was promising in terms of accuracy (F1=0.81). Moreover, the results show that applying LR, SVM, and RF using unigram language model is significantly better than using the bigram language representation.