Transfer learning have recently proven to be very powerful in diverse Natural language processing (NLP) tasks such as Machine translation, Sentiment Analysis, Question/Answering. In this work, we investigate the use of transfer learning (TL) in Dialectal Arabic sentiment classification. Our main objective is to enhance the performance of Sentiment classification and overcome the low resource issue of Arabic dialect. To this end, we use Bidirectional Encoder Representation from Transformers (BERT) to transfer contextual knowledge learned from language modelling task to sentiment classification. We particularly use the multilingual models mBert and XLM-Roberta, the specific Arabic models ARABERT, MARBERT, QARIB, CAMEL and the specific Moroccan dialect Darijabert. After carrying out downstream fine-tuning experiments using different Moroccan SA datasets, we found that using TL significantly increase the performance of sentiment classification in Moroccan Arabic. Nevertheless, though specific Arabic models have proven to perform much better than multilingual and dialectal models, our experiments have demonstrated that multilingual models can be more effective in texts characterized by an extensive use of code-switching.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.