Automating customer feedback analysis in E-commerce: A multi-Model approach

Understanding customer satisfaction in e-commerce is crucial for businesses to remain competitive. While traditional feedback analysis methods are labour-intensive and subjective, machine learning advances have enabled more efficient and scalable sentiment analysis. However, existing models struggle with aspect-based sentiment analysis (ABSA), particularly in detecting implicit aspects and handling mixed sentiments. This paper presents a multi-model machine learning pipeline designed to enhance ABSA by integrating fine-tuned Large Language Models (LLMs) with BERT and RoBERTa-based models. The pipeline consists of an LLM-generated synthesized annotated feedback model, a BERT-based aspect detection model, a RoBERTa-based ABSA model, and an LLM-based ABSA model for handling implicit aspects and mixed sentiments. Additionally, a RoBERTa-based model is employed for overall sentiment detection. By leveraging both manually annotated and synthetic data, the pipeline improves sentiment classification accuracy and aspect coverage, even in data-scarce environments. The results demonstrate that combining multiple models enhances detection accuracy compared to single-model approaches. This study provides a scalable and effective solution for e-commerce feedback analysis, offering businesses valuable insights for improving customer experience and decision-making.