Comparison of BERT Models and Machine Learning Methods for Sentiment Analysis on Turkish Tweets
Zekeriya Anıl Güven
Abstract
Users can freely express their opinions about many events on social media platforms. It may be necessary to analyze the data in order to get the opinion of the society about these events. Therefore, sentiment analysis studies are gaining importance today. Many different methods and models are used for sentiment analysis. While language models such as the BERT model are widely used in the English language, there are very few studies for the Turkish language in sentiment analysis. In this study, sentiment analysis was performed on tweets using BERT models and machine learning methods. In addition, the trained BERT models and machine learning methods were compared. Among the Random Forest, Naive Bayes and Logistic Regression machine learning methods, Logistic Regression was the most successful method with 98.4%. BERT models achieved 98.75% accuracy and surpassed the success of machine learning methods. The positive effect of the BERT model on sentiment analysis was shown with this study.