Simple definition
An embedding is a numerical vector representation of features from text, images, or other data.
Purpose and use
An embedding represents data as a learned or calculated numerical vector for selected comparison or processing tasks.
How does it work?
The model places items in a space where proximity can reflect relationships useful to the task.
Example and verification
For semantic search, verify consistency between the model indexing documents and the model representing queries.
Points to watch
Vectors from different models are generally not directly comparable. Proximity guarantees neither identical meaning nor factual accuracy.
In short
An embedding depends on a model and metric and should be evaluated for intended uses.