Deep Bidirectional LSTM Network Learning-Based Sentiment Analysis for Tunisian Dialectical Facebook Content During the Spread of the Coronavirus Pandemic.

Sentiment analysis (SA) is a multidisciplinary field that aims to predict sentiment tone or attitude expressed in a text, SA using social media data has become a popular topic especially during critical events such as natural disasters, social movements and recently the spread of the Coronavirus Pandemic. Sentiments can be expressed explicitly or implicitly in text and identifying these expressions can be challenging. SA in Tunisian dialect is particularly difficult due to the complexity of the language, its morphological richness and the lack of contextual information. Recently, deep learning (DL) models have been widely adopted in the field of SA, especially in the context of Arabic SA. These models, such as Bi-directional LSTM networks (Bi-LSTM) and LSTM networks, have shown to achieve high accuracy levels in sentiment classification tasks for Arabic and dialectical text. Despite the successes of DL models in Arabic SA, there are still areas for improvement in terms of contextual information and implicit mining expressed in different real-world cases. In this paper, the authors introduce a deep Bi-LSTM network to ameliorate Tunisian SA during the spread of the Coronavirus Pandemic. The experimental results on Tunisian benchmark SA dataset demonstrate that our model achieves significant improvements over the state-of-art DL models and the baseline traditional machine learning (ML) methods. We believe that this contribution will benefit anyone working on Tunisian pandemic management or doing comparative work between Tunisian and other jurisdictions, which can provide valuable insights into how the public is responding to the crisis and help guide pandemic management decisions.