
This project showcases how I used BERTopic, a transformer-based topic modeling approach, to extract meaningful themes from a noisy dataset of product reviews. By leveraging contextual embeddings and cosine similarity, I clustered similar comments and visualized them in three interactive charts—a bubble map, bar chart, and heatmap—built with Plotly. To enhance interpretability, I integrated GPT to auto-label the topics based on keywords and sample excerpts. The result is a fully automated pipeline that turns unstructured text into clear, explorable insights.
Other visualizations:


Learn more from the blog post.