Comparative Study of Gemini-Based Semantic Clustering and LDA Topic Modeling: Analyzing Wellness Product Reviews on Amazon.com

This comparative study evaluates semantic embedding-based clustering using Gemini against traditional Latent Dirichlet Allocation (LDA) topic modeling across 1,998 verified Amazon reviews from four leading wellness meal-replacement products, including Garden of Life. Review data were collected via Python Selenium and BeautifulSoup, preprocessed, and analyzed using both models. Gemini generated nine semantically distinct clusters per product, while LDA extracted five probabilistic topics. Quantitative assessment across three dimensions reveals that Gemini achieved a topic coherence score higher than LDA (0.452 vs. 0.346) and a topic diversity score 1.64 higher, reflecting superior semantic cohesion and reduced keyword redundancy. Conversely, LDA slightly outperformed Gemini in silhouette score, suggesting more structurally compact clusters. These findings underscore the complementary strengths of contextual embedding models and frequency-based topic models. We propose a hybrid workflow leveraging Gemini for nuanced topic interpretation and LDA for cluster structure optimization, offering a scalable and interpretable approach to large-scale e-commerce review analysis.