clustering

Clustering groups similar items by embedding them into a vector space and applying unsupervised algorithms like KMeans, DBSCAN, or HDBSCAN. Provide your content, choose an embedding model and clustering method, and the Skill assigns items to cohesive clusters with automatically named topics.

You need an embedding model to convert items into vectors, a Milvus vector store to index them, and a clustering algorithm. The Skill handles the rest—storing embeddings, computing similarities, and partitioning data into labeled clusters.

Yes. Unsupervised clustering reveals hidden patterns and topics in text, product descriptions, or user data without labeled training data. The Skill automatically assigns human-readable labels to clusters so you understand what each group represents.

Clustering organizes data into semantic groups, making it easy to compare compositions, spot outliers, and identify unexpected patterns. This guides downstream tasks like segmentation, recommendation, or deeper investigation of unusual items.

KMeans partitions data into a fixed number of clusters; DBSCAN finds density-based groups and flags outliers; HDBSCAN is a hierarchical variant that adapts to varying cluster densities. Choose based on whether you know cluster count and how strict density requirements are.

The embedding model handles most normalization. Provide clean, deduplicated items in supported formats; the Skill converts them to vectors and applies clustering. Minimal preprocessing is needed when using modern embedding models.