Emoji translation for sentiment analysis in Algerian Arabic dialect

Sentiment analysis (SA) is an important natural language processing (NLP) field that involves extracting sentiments and opinions from text data. Although SA has advanced significantly, its application to dialectal Arabic text presents challenges due to linguistic nuances and resource constraints. This research investigates the incorporation of emojis into SA for Algerian Arabic dialect (AAD), marking the first exploration of its kind in this area. Specifically, we focus on emoji translation, building upon prior studies highlighting emojis, potential in SA and their translation into meaningful words or sentences as a preprocessing approach. We evaluate the impact of this approach on enhancing sentiment classification in AAD text, specifically focusing on customer reviews of Algerian telephone operators. After preprocessing, including various emoji translation techniques, we employ transfer learning by fine-tuning DziriBERT model on a compiled Algerian dialect dataset. Our results demonstrate promising outcomes and offer novel conclusions and perspectives in AAD sentiment analysis.