Exploring Word Embedding for Arabic Sentiment Analysis

Abstract

In Natural Language Processing (NLP), the manual features (part-of-speech tagging, stemming…) might not be helpful sometimes to deciding the feeling expressed in a sentence. That More properties need to be considered. Word embeddings which are the key component for learning the text features, has just started to appear in Arabic sentiment analysis. On the other hand, Deep Neural Networks were widely used recently for this task, especially for the English language. In this paper, we focus on the Tunisian dialect sentiment analysis used on social media using a Convolutional Neural Networks and Bidirectional Long Short-Term Memory. The results show that our models on the publicly available TUNIZI dataset achieved superior performances than the other models applied for the same dataset.