What Are Embeddings and Why They’re Incredibly Useful

1 minute read


If you’ve heard of the term “embeddings” before but don’t quite understand it, this post is for you. I’ll provide an intuitive explanation of embeddings and their significance.

What are embeddings?

Embeddings are a numerical representation of content like text, images, documents, and audio. These numbers or vectors represent the semantic meaning of the content as coordinates or vectors. Think of these vectors as points in a highly-dimensional (and incomprehensible) space - far beyond the 2D and 3D spaces we’re comfortable visualising. Vector embeddings can have 384, 768 or 1536 dimensions.

No matter the length of the content, embeddings generated by the same embedding model will always have the same length. The term embedding is used interchangeably with text embedding and vector embedding.

Why are embeddings useful?

Mapping the meaning of content into a spatial representation is powerful for many tasks. By quantifying the meaning of text into specific points in space, we can measure how similar or different two pieces of text are by calculating the distance between their vectors. The most popular method for this is cosine similarity.

This approach overcomes the limitations of exact-match searches, enabling semantic or “vibes-based” search that use the underlying meaning of content.

An embedded dataset enables algorithms to quickly search, rank, group, and more. Here are some practical applications framed in terms of text similarity:

  • Search: How similar is a query to a document in a database?
  • Spam detection: How similar is an email to existing spam examples?
  • Chatbot: How similar is the user’s message to known intents?

Embeddings are essential in search engines, recommendation systems, and chatbots.

Their ability to quantify semantic meaning makes them invaluable across various industries, improving our ability to process and understand large unstructured datasets.