Similarity and Distance

1. What does cosine similarity primarily measure in vector embeddings?

The length of the vectors The angle between two vectors The magnitude of the vectors The distance between two points

2. In SQL Server 2025, how is cosine distance calculated?

VECTOR_DISTANCE('cosine', a, b) = cosine similarity VECTOR_DISTANCE('cosine', a, b) = 1 - cosine similarity VECTOR_DISTANCE('cosine', a, b) = cosine similarity + 1 VECTOR_DISTANCE('cosine', a, b) = cosine similarity - 1

3. What is the purpose of setting a distance threshold in semantic search?

To increase the number of results returned To ensure all results are relevant To filter out irrelevant matches To minimize processing time

4. Why is it important that vectors for cosine distance comparisons come from the same model?

Different models produce vectors of different lengths Vectors from different models have incompatible geometry Cosine distance calculations require identical dimensions Different models lead to irrelevant embeddings

Continue: Embeddings in Practice → Review this lesson Retake quiz