Classifying and Predicting The Rating Sentiment of Women's E-commerce Clothing Reviews: A Comparative Study Using SVM, ANN, and BERT Models

Product reviews are crucial for both customers and sellers in determining the quality of certain products. Alongside the review, the rating is also provided to give insight into the overall point of the product. As the review often becomes a consideration for a customer when buying a product online, the review needs to give an objective viewpoint. Therefore, it is important to determine whether the review is reliable. This could be seen both from the rating given by customers and the sentiment of the review. Machine learning and deep learning methods can be applied to classify the review sentiment effectively. The main objective was to check whether the review was considered positive or negative. The review was provided from women’s clothing e-commerce by Nicapotato, which is available on Kaggle. This paper used both SVM and ANN for machine learning methods and BERT for deep learning method to determine the best method for the given dataset by identifying the most reliable and accurate model for classifying reviews. For the text vectorization, SVM and ANN models used TF-IDF technique while BERT model used sentenceBERT. The best method was determined quantitatively by comparing each model's accuracy and F1 Score results. Based on the accuracy and F1 Score result, it turned out that using TF-IDF for text vectorization with ANN gave the best performance compared to other models.