Embeddings map

Embeddings place words and sentences at points in space so that similar meanings sit close together. This map uses real embeddings from an open model.

What to notice

• “dog” is nearest “puppy” and “cat”: similar things are used in similar ways, so they sit together.
• “How do I get my money back?” lands next to “refund” and “Refund policy”, even though they share almost no words. That is how search by meaning works.
• “bank” is nearest “river” before it reaches money words. A single word can have several meanings, and its embedding blends them. Context fixes this inside a model.

Each item was embedded with the open model all-MiniLM-L6-v2 (384 numbers per item). The neighbours and similarity scores are measured in all 384 dimensions. The 2D positions are a projection for drawing, so distances on the map are only approximate: trust the neighbour list over the picture.