Aspect-Based Sentiment Analysis of Clothing Product Reviews by Transformer Models

Abstract

This study analyses user reviews of clothing products on e-commerce platforms through aspect-based sentiment analysis. Aspect-based sentiment analysis is critical for improving customer satisfaction in the clothing sector and helping customers make more informed product decisions. Due to the inadequacy of the existing Turkish clothing review datasets, a new dataset containing customer comments and the most evaluated product features of various clothing categories was created using the web scraping. Using the created dataset, transformer-based language models such as BERT, DistilBERT, and GPT-2 were fine-tuned according to the aspect-based sentiment analysis. The performance of fine-tuned models was evaluated with the test dataset. The test results show that the BERT model exhibits superior performance in metrics such as accuracy and F1 score compared to other models. These results support that aspect-based sentiment analysis is an effective tool for evaluating customer reviews.