Applying deep learning models for multi-label analysis of Turkish E-commerce comments

E-commerce stands out as one of the rapidly growing sectors in the increasingly digital world. One of the most important factors influencing consumers' online shopping decisions is customer reviews, where users share their experiences. However, manually analyzing these reviews is a time-consuming and costly process for large datasets. This study aims to process customer reviews obtained from e-commerce platforms quickly and accurately using artificial intelligence-based classification methods. Customer reviews collected from leading e-commerce platforms in Turkey have been categorized into multi-label classification categories such as "Price-Performance", "Returned", "Good Packaging, Fast Delivery", "Low Quality, Defective, Bad Packaging", and "Recommended, Quality Products". These categories have contributed significantly to the field of sentiment analysis by reflecting not only the content of the reviews but also the positive or negative emotional states of the customers. Microsoft SQL Server was used for data storage and management, and data scraping was carried out using Selenium Web Driver and Html Agility Pack tools. The reviews were processed with TF-IDF representation in natural language processing tasks and classified using deep learning models based on artificial neural networks, such as LSTM, CNN, GRU, RNN, BiLSTM, and BERT, as well as traditional machine learning algorithms like Logistic Regression and Random Forest. Performance evaluation metrics such as accuracy, sensitivity, precision, F1-score, log loss, and confusion matrix were used. Notably, the BERT model outperformed other methods in large datasets, demonstrating superior performance in classification and context extraction due to its contextual understanding capabilities. BERT's transformer-based architecture, with its ability to analyze language context bidirectionally, allowed for a deeper understanding of text meaning and more accurate and meaningful classifications. The artificial intelligence models used in this study achieved high accuracy rates in classifying customer reviews, and it was demonstrated that the developed AI-supported system guided consumers in making informed decisions and contributed to improving marketing and product strategies for businesses. BERT's superior performance, particularly in multi-label classification tasks, stands out for its success in understanding language context accurately. In this context, the study provides an innovative and effective solution for analyzing customer reviews in the e-commerce sector.