MTEB Clustering: MTEB Clustering Task Group

How It Works

Clustering scores measure whether documents from the same topic end up grouped together when the model's embeddings are clustered. The metric (V-Measure) does not care about specific cluster labels — only that documents about the same thing land in the same bucket.

Embeddings are run through k-means with the known number of clusters, then scored against the ground-truth grouping using V-Measure. Per-dataset scores are averaged.