Customer Feedback Sentiment Analysis in E-Commerce Using Hybrid Approaches: Combining LLMs and Knowledge Graphs

In the competitive landscape of ecommerce, customer feedback plays a crucial role in shaping service improvement and purchasing decisions. However, traditional sentiment analysis techniques often fall short in capturing contextual nuances, sarcasm, or domain-specific expressions. This research addresses these limitations by proposing a hybrid sentiment analysis framework that integrates Large Language Models (LLMs) and Knowledge Graphs (KGs) using a Retrieval-Augmented Generation (RAG) approach. The urgency of this research lies in the need for more accurate and explainable sentiment analysis in platforms like Tokopedia, which host vast volumes of customer reviews. The study uses a curated dataset of 1440 Tokopedia customer reviews and applies standard preprocessing techniques including tokenization, stemming, lemmatization, and stopword removal. Tools such as DeepSeek, GPT-4, and domain-specific knowledge graphs are utilized to encode semantic context and enrich information retrieval. The proposed hybrid model combines vector-based and graph-based retrieval to enhance classification accuracy. Evaluation metrics—precision, recall, and F1-score—show that the model achieves high contextual precision (up to 88.21%), while context recall ranges widely ($33.33 \%- 100$%), indicating variability in retrieval completeness. These results demonstrate that combining LLMs with structured knowledge can significantly improve the relevance and reliability of sentiment analysis in ecommerce applications.