Vector Search indexes
Now that you’ve successfully generated vector embeddings for your movie plots, you need a way to efficiently compare these vectors to find similar content. But how do you efficiently search through vectors with hundreds or thousands of dimensions? This is where Atlas Vector Search indexes come into play.
Understanding the technology behind Vector Search
When we work with high-dimensional vector data, traditional search methods aren’t efficient. MongoDB Atlas uses hierarchical navigable small world (HNSW) graphs to power Vector Search. But what exactly is HNSW, and why is it so effective?
The evolution to HNSW
HNSW evolved from two important data structures:
- Skip lists: These are sorted linked lists with multiple layers, each containing list points. This structure increases search efficiency.
Figure 9.7: Skip list layers for navigating to value 9
- Navigable small world (NSW) graphs: Points (vertices...
