Employing Explainable Artificial Intelligence For Adapting Multi-Dialectical Arabic Textin Decision-Making Scenarios

Abstract

As machine learning has gained prevalence across multiple fields, its complexity, as well as that of deep learning (DL), has continuously increased. Over the last span of years, the use of deep learning across several fields in the aim of advancing, improving and accelerating the classification in general or improving the sentiment analysis in specific, came with it a lot of ambiguity of how these algorithms works for non-experts. which in return researchers started to implement the XAI Algorithms in order to proof and clarify the basic process of how the sentiments classified. While most preexistent studies have utilized the XAI across English and other Latin-based languages for extended reasons, this particular study was utilized to attempt explaining Attention-based long short-term memory results used across the process of Arabic multi-dialect dataset. With the use of local interpretable model-agnostic explanations, our objective was to further demonstrate and simplify the LSTM-led prediction of sentiment polarity process. We attempt to explain how the outcome of attention-based LSTM reaches that of sentiment analysis by applying XAI, thereby yielding potential insights into the study of complex DL models across domains. And finalizing it with a simplified comparison of the probabilistic weights represented on the shown examples with the occurrence of words across the dataset.