This project clusters products by their titles and assigns topics. Initially using BERT, PCA, and t-SNE, the results were noisy. The improved approach with SBERT, UMAP, and HDBSCAN provides clearer clusters. Topics are assigned using Llama-3-8b.
unsupervised-learning umap dbscan-clustering hdbscan sentence-bert content-clustering llm-inference llama3-8b
-
Updated
Jul 11, 2024 - Jupyter Notebook