Multilingual and Multi-Class Sentiment Classification Using Machine Learning, BERT, and GPT-4o-mini

In this study, we investigate multilingual and multiclass sentiment classification by analyzing datasets in Turkish, English, and Italian. The proposed approach consists of three main stages: sentence representation extraction, classification, and performance evaluation. First, sentence representations were extracted from these datasets using the distiluse-base-multilingual-cased-v1, sentence-transformers/LaBSE, and Alibaba-NLP/gte-multilingual-base models. These representations were then used as input for Logistic Regression (LR), Support Vector Machines (SVM), Random Forest (RF), and Naive Bayes (NB). Additionally, fine-tuned BERT-base-multilingual-cased and GPT-4o-mini were directly employed as end-to-end models. The classification performance of the models was evaluated using accuracy, F1-score, precision, and recall. Additionally, a confusion matrix analysis was conducted for each dataset to examine classification performance in detail. The experimental results show the influence of embedding models and learning algorithms on multilingual multi-class sentiment classification.