Reviews Classification using BERT

In this project, I fine-tuned a Bert-based model to classify Talabate company reviews. The dataset was in

Arabic. My job was not just classifying reviews into negative and positive but to classify negative reviews into categories to help decision-makers improve the company’s services. I have to classify some samples of the available dataset manually to fine-tune bert based model which was trained on Arabic text. The model was downloaded from the Hugging Face website. Only 2500 samples used for training and validation accuracy reached 74% on test data which was acceptable performance for our project and the available data.