Sentiment Analysis for Low-Resource and Arabic Dialects: A Comprehensive Review
Taoufik Amzil, Ayoub Jannani, Nawal Sael · 2025
The paper presents a comprehensive review of sentiment analysis (SA) across low-resource languages and Arabic dialects. While SA models have been successful for languages with abundant resources, their application for low-resource languages remains challenging due to lack of labeled data and intricate linguistic characteristics. We examine 30 research papers to reveal investigated languages, employed datasets, and proposed methodologies. Our analysis indicates that transformer-based models, particularly those specifically adapted for dialects like AraBERT and DarijaBERT, generally achieve better results than traditional machine learning and deep learning methods. The performance of these models varies considerably depending on the quality of the data, the level of context awareness, and the model's specificity to a particular dialect. Combined methods that integrate transformers with recurrent or convolutional networks also show encouraging results, often exceeding 85% accuracy. However, challenges remain, including the shortage of data, inconsistent ways of evaluating performance, and limited investigation into multimodal and transfer learning strategies. We conclude by suggesting future research directions that aim to address these limitations and enhance SA for languages with fewer resources.