Embeddings are mathematical representations of phrases that capture the semantic meaning as vector of numerical values.
Phrases with similar semantic meanings will have similar vector representations => the distance between these vectors is going to be small.
One common use of embeddings is Semantic Similarity Search – If I have knowledge library consisting of phrases and I receive a question from a user, I can locate the most relevant information in my library by finding the data that is closer to the user’s query.
In contrast, Full Text Search only returns exact matches.