Sentiment Analysis in Maghrebi Arabic Dialects with Enhanced BERT Models and Big Data Processing

Sentiment analysis in Maghrebi Arabic presents significant linguistic difficulties because of resource constraints. Among these difficulties are the several dialects spoken in the Maghrebi area and the peculiar grammatical systems. We introduce a complete framework that uses the Bidirectional Encoder Representations from Transformers (BERT)-mini and BERT-base models, improved with large data processing capabilities offered by Apache Spark, to improve scalability and performance. Selecting a variety of datasets, sophisticated preprocessing methods like tokenization, normalization, and keeping special characters and emojis are used, and several model architectures are carefully explored. Our method works; BERT-mini reaches accuracy rates of up to 0.885 and BERT-base reaches 0.899. Conventional machine learning techniques are much outperformed by these outcomes. This paper shows how well BERT models may be integrated with big data technology and points up areas that need more investigation. Our framework offers important insights and useful applications in many domains, including social media monitoring, market analysis, and customer feedback evaluation. It does this by tackling the linguistic complexity of Maghrebi Arabic.