Machine Learning-Based Irony Detection and Sentiment Analysis in Arabic Tweets with Emojis

Social media platforms depend mainly on written texts for communication. Nevertheless, such text regularly lacks the emotional cues that are present in face-to-face encounters, which can result in ambiguity, given how informal and short social media messages are. Irony is a form of communication where the intended meaning is opposed to the literal meaning and detecting it in these online messages could be challenging. This challenge is more relevant in Arabic texts due to the Arabic language’s rich morphology, cultural nuances and various dialects. Despite being one of the most spoken languages globally, Arabic is still underrepresented in Natural Language Processing (NLP) research, particularly in the domain of irony detection and sentiment analysis which is more widely researched in English. In addition, the increased usage of emojis in online platforms adds another challenge as they represent both emotional cues and ironic meanings. The majority of the existing approaches of irony detection and sentiment analysis mostly rely on text, ignoring the important emotional and contextual information provided by emoji. This study attempts to address these issues by focusing on irony detection and sentiment analysis in Arabic tweets incorporating emojis. The research uses the “ArSarcasMoji” dataset, and we employ a CNN model for irony detection alongside machine learning classifiers (Naïve Bayes, Decision Tree, and Random Forest) for sentiment analysis. Our results show that CNN effectively identifies irony, while Naïve Bayes outperforms the other two classifiers in sentiment analysis. Additionally, we compare our two-step approach with a transformer-based BERT model that jointly handles irony detection and sentiment analysis. While BERT achieves slightly higher results, our approach remains a simple, efficient, and lightweight alternative. This work implements a small web application allowing the user to classify irony and analyze sentiment of Arabic tweets containing emojis with the proposed solution.