Clusteringconcept

Grouping items so similar ones share a cluster — in knowledge compilation, grouping embeddings by similarity into topic clusters.

Clustering turns a flat embedding set into navigable topic groups (the cluster graph view), a cheap runtime structure.

Where it appears

Descend

Historical evolution

Connected-components clustering on a similarity graph (threshold >= 0.1) with centroid averaging.

Implementation details

SKCE's concept graph is edge-weighted by co-occurrence rather than geometric clustering, but the principle is shared.

Sources

View in graph →