Automated Categorization of Turkish E-commerce Product Reviews Using BERTurk

Thanks to rapid technological developments, users can now easily access and distribute information. While users have the opportunity to share content and information as they wish, they can also use the information shared to improve themselves in line with their own interests. Especially when buying a product, they make their final decision about the product by examining the content of the comments made about the product. Automatically classifying the comments made about the product by the system and displaying them in the relevant category is one of the issues that has been studied recently. In this study, comments on e-commerce sites were automatically categorized. The dataset was created by collecting comments in Turkish about phones, computers and headphones produced by the same company on Amazon.com.tr, one of the world’s largest e-commerce platforms, with the help of a Python script. User comments were automatically sorted into categories by machine learning algorithms such as Naive Bayes, Linear Support Vector Classifier and Random Forest algorithms and pre-trained and fine-tuned Bert Multilingual and BERTurk models based on transfomer architecture. The results obtained were compared with the help of the F1-score metric. When the results of different machine learning algorithms and BERT models were compared, it was seen that the BERTurk model gave more accurate results than other models.