Deep Attentional Bidirectional LSTM for Arabic Sentiment Analysis In Twitter

Abstract

Along with the emergence of user reviews, emotions, feedback, and opinions in social networks towards a specific topic, product, event, or such a service. Sentiment Analysis has recently considered one of the fundamental research areas which lie at the intersection of numerous fields of research include data mining, computational linguistics, and Natural Language Processing (NLP). It concerns classifying a given piece of writing into sentiment polarity. Furthermore, deep learning has shown rich data modelling abilities to deal with complex and large datasets, in addition, it is recognized as the state-of-the-art based approach for different research fields. Although, the current state-of-the-art sentiment analysis tailored to the Arabic language still needs improvements because of its morphological richness, ambiguity, and lack of its resources. To advance this task, a novel Attentional Bidirectional LSTM architecture was proposed in order to determine richer semantic information and to extract the contextual information in both directions. We also investigated the effect of the word2vec pre-trained model to produce the word embeddings representation and to capture semantic information from Arabic tweets. To validate the performance of the proposed architecture, we assessed it in a holistic setting across three benchmark Arabic sentiment tweets datasets. Thus, the experimental results demonstrate that the proposed architecture outperforms the current state-of-the-art deep learning-based methods. Besides, it performs well compared with the baseline classical machine learning methods.