3D Name Clustering

At Namey, we love finding new ways to explore the relationships between names. Our previous blog posts have shown name clusters as 2D scatter plots. Today we're taking that a step further by visualising the same clusters in three dimensions, revealing structure that a flat projection necessarily hides.

The 3D visualisation below uses the same enhanced embeddings as our enhanced clustering model. We've applied t-SNE with three components instead of two, then used K-means with 200 clusters to colour the points. Drag to rotate, scroll to zoom, and click any point to open that name in the Namey app.

In three dimensions you can see clusters that appear merged in 2D actually occupy distinct regions of space. Names with shared cultural heritage — Celtic names, classical names, modern invented names — tend to group together even when their 2D projections overlap with unrelated clusters.

How it works

Each name is represented by a 1,536-dimensional enhanced embedding that captures phonetic patterns, spelling, cultural context, and list membership. t-SNE reduces these to three dimensions while preserving local neighbourhood structure. K-means then assigns each name to one of 200 clusters, which determines its colour in the visualisation.

The 3D layout is rendered using Three.js via React Three Fiber. All 100,000+ names are rendered as a single GPU draw call using a point cloud, keeping interaction smooth even on modest hardware.