Definition
A mathematical metric used in vector databases to measure the semantic dissimilarity between two high-dimensional embeddings based on the angle between their vectors, typically calculated as the inverse of cosine similarity. In RAG architectures, it prioritizes the thematic orientation of data over the magnitude of the vector, ensuring that document length does not disproportionately skew retrieval results.
Focuses on the direction of semantic meaning rather than the Euclidean 'straight-line' distance between points.
"Two searchlights in a dark sky: the angular distance is the width of the gap between the beams, indicating how closely they point to the same star, regardless of how far the light reaches."
- Cosine Similarity(Inverse Metric)
- Vector Embedding(Prerequisite)
- Euclidean Distance (L2)(Alternative Metric)
- Dot Product(Mathematical Component)
Conceptual Overview
A mathematical metric used in vector databases to measure the semantic dissimilarity between two high-dimensional embeddings based on the angle between their vectors, typically calculated as the inverse of cosine similarity. In RAG architectures, it prioritizes the thematic orientation of data over the magnitude of the vector, ensuring that document length does not disproportionately skew retrieval results.
Disambiguation
Focuses on the direction of semantic meaning rather than the Euclidean 'straight-line' distance between points.
Visual Analog
Two searchlights in a dark sky: the angular distance is the width of the gap between the beams, indicating how closely they point to the same star, regardless of how far the light reaches.