Leila ChekhchoukhAsma Gaouer2025-11-092025MM/898https://dspace.univ-bba.dz/handle/123456789/973This study addresses the scarcity of resources in Natural Language Processing (NLP) for the Algerian dialect, particularly in the context of scientific discourse. We have constructed a sentiment-labeled dataset comprising sentences related to scientific research, annotated across three languages: English, Modern Standard Arabic, and Algerian Arabic. Each sentence is tagged with its corresponding sentiment polarity (positive, negative, or neutral). The dataset aims to facilitate the development of domain-specific sentiment analysis models tailored to the Algerian dialect. Preliminary experiments utilizing transformer-based models, such as BERT variants fine-tuned on this dataset, demonstrate promising results in accurately classifying sen timent within this under-resourced dialect. This work contributes to the advancement of NLP tools for the Algerian dialect and underscores the importance of creating specialized resources for low-resource languages.enThesis