Advancements in Arabic Tweets Sarcasm Detection & Sentiment Analysis with Fine-Tuned Models

Abstract

This research paper addresses the significant challenge of sarcasm detection and sentiment analysis within the context of Arabic tweets. It highlights the efficacy of the “AraBert-Arabic-Sentiment-Analysis” model, a pre-trained model that demonstrates robust performance in these tasks. When fine-tuned, this model achieves an accuracy of 87.8% in sarcasm detection and 77.4% in sentiment analysis. Our findings also suggest that extensive preprocessing of data is not invariably beneficial. The optimal results were observed with minimal preprocessing, underscoring the importance of retaining critical features in the data. This study emphasizes the need for a balanced approach to data preprocessing to enhance the effectiveness of computational models in handling nuanced linguistic tasks.