Vector Similarity
Definition
Vector similarity measures how close two vector representations are according to a chosen metric.
For embeddings, cosine similarity can compare the direction of vectors; other metrics consider, for example, distance or dot product. The meaning depends on the model and the task.
Simply explained
Like comparing two positions on a semantic map.
Why it matters in practice
Semantic search systems can find relevant content without identical words.
Example from business
Hypothetical example: A search associates "pay invoice" with a document about payments.
Limits and pitfalls
- High similarity neither proves truth nor identical meaning in every context; different metrics are not arbitrarily interchangeable.
Origin & history
Vector comparisons originate from mathematics and information retrieval and are applied to learned representations.
Use in mid-sized companies
- Document search, similar products, and duplicate candidates.
Frequently asked questions
Is a similarity value a probability?
Not automatically; the metric used must be explained.
Sources
- cosine_similarity · scikit-learn: cosine_similarity –
Editorially reviewed on: