Vector Similarity
A technique that encodes text into high-dimensional vectors and measures semantic closeness by computing distances between vectors, such as cosine distance. It is one of the core mechanisms of the retrieval stage in RAG.
Core Facts
Timeline (last 90 days)
In a standard RAG pipeline, vector similarity does not equal context accuracy, and the retriever may pull outdated documents or documents at the wrong application level.
Semantic caching reuses existing answers by comparing the mathematical similarity of prompt vectors, bypassing duplicate upstream model calls, but false hits may return incorrect answers.
RRF uses only ranks rather than raw scores, avoiding the problem that the dimensions of dense vectors and sparse BM25 are not comparable.
Vector similarity measures semantic proximity, not temporal validity or whether information has been superseded.
All Facts (5)
Graph databases store entities and relationships in the form of nodes and edges, and natively support multi-hop reasoning.
75%UnverifiedVector similarity measures semantic proximity, not temporal validity or whether information has been superseded.
60%UnverifiedIn a standard RAG pipeline, vector similarity does not equal context accuracy, and the retriever may pull outdated documents or documents at the wrong application level.
50%UnverifiedSemantic caching reuses existing answers by comparing the mathematical similarity of prompt vectors, bypassing duplicate upstream model calls, but false hits may return incorrect answers.
50%UnverifiedRRF uses only ranks rather than raw scores, avoiding the problem that the dimensions of dense vectors and sparse BM25 are not comparable.
50%