Back to Home
Embeddings & Search
Clustering
Clustering means grouping similar items together, without being told in advance which items belong. Gmail uses it to group your emails into categories. Spotify uses it to build genre playlists. AI uses it to discover patterns in data automatically.
Known words use preset semantic positions. Unknown words get a character-based estimate.
Interactive Playground
3
Statistics
12
Total Items
3
Clusters
Largest Cluster
Smallest Cluster
How K-Means Works
1
Pick K random centres
K = number of groups you want (try changing the slider!).
2
Assign each item to nearest centre
Measure distance, closest centre wins.
3
Move centres to group midpoints
Recalculate the average position for each group.
↺
Repeat steps 2 & 3
Until nothing changes, groups are stable.
Live Visualization
2D Map (D1 = X · D2 = Y · color = cluster)
Cluster Groups
🎓
Tips
3 tips
K = the number of groups you decide. Try the slider, fewer clusters merge topics together; more clusters split them apart.
Clustering is unsupervised, you don't label anything in advance. The AI discovers groups purely from data.
K-Means starts with random centres, so running it again can give slightly different clusters. That's normal.
💡
Key Takeaway
Clustering helps organize many items into smaller similar groups without needing labels or examples in advance.