Investigation of Local Interpretable Model-Agnostic Explanations (LIME) Framework with Multi-Dialect Arabic Text Sentiment Classification

Sentiment Analysis (SA) aims at determining the positive, negative, or neutral polarity in unstructured data, which is mostly social media data such as tweets, blogs, or online reviews. Various Machine Learning (ML) algorithms are used for social text polarity classification, however without providing sufficient explanation for the obtained decisions. Explainable Artificial Intelligence (XAI) is an emergent field of AI that helps providing more visibility and explanations that leads to more understanding of the ML models’ output. This paper presents an investigation study on the effectiveness of well-known supervised ML algorithms for sentiment classification of Arabic social media text (tweets in the Arabic language) concerning different topics. That has been achieved through utilizing the Interpretable Model-Agnostic Explanations (LIME) XAI framework to attain reasonable explanation of the models’ performance based on highlighting the features that the model considered necessary while making the decision. Two benchmark publicly available social media text datasets of Modern Standard Arabic (MSA) and dialectal Arabic are used to evaluate the performance of the classifiers. Experimental results show that the implemented ML models trained using a Random Forest (RF) classifier with the Hotel Arabic Reviews dataset (HARD) dataset attains a significant accuracy of 93.40 % and F-score values of 96 % and 97.20 %, considering positive and negative polarity sentences, respectively.