6 Comments
User's avatar
Vivek Somvanshi's avatar

Very Helpful Article

Expand full comment
Paul Iusztin's avatar

Thanks, man 🥂

Expand full comment
Vijendra Rajput's avatar

Thanks for sharing insights

Expand full comment
Paul Iusztin's avatar

My pleasure 😇

Expand full comment
KimSia Sim's avatar

I understand that there are three kind of embeddings

Token-based, tree-based, graph-based

This article explainer covers token-based or all 3?

Expand full comment
Paul Iusztin's avatar

In the end, all embeddings mean the same thing: "they encode some information into a vector"

Depending on your data and the method you use, you can categorize them as token-based, tree-based, graph-based, image-based, etc.

For example, you can encode text data in all the ways you proposed, but depending on the algorithm, you emphasize different aspects. If you use a graph-based (GNN) encoder, your embeddings will contain much more information on the relationships between different entities in your database.

Hope that helps!

Expand full comment