BERTurk-Based Sentiment Analysis on E-Commerce Multi Domain Product Reviews

Product reviews on e-commerce platforms constitute an important source of information for customers' shopping processes. Learning about the various features of products and evaluating user experiences makes shopping more reliable and provides sellers with valuable feedback on customer satisfaction. In order for sellers to make strategic decisions about their products, customer satisfaction and product feedback should be analyzed in detail. For this purpose, sentiment analysis methods are applied on the data to analyze the sentiment of the comments. In our study, sentiment analysis was performed using comments from the Trendyol e-commerce site. Our dataset was studied on a total of 73.398 data by extracting data from six different categories, namely Computer, Phone, Shoes, Clothing, Cosmetics, Sports and Outdoors, via Selenium. This dataset was divided into 80% training data and 20% test data.. The training set was validated with the fold cross-validation method. As a result of the experiments, among the traditional machine learning models, Support Vector Machines (SVM) achieved the highest accuracy rate with 88%, while the BERT model was determined as the most successful model with an accuracy rate of 95%.