Word Map

See how closely AI relates words to each country.

About

This project was created by Ubaada. The code is available on GitHub.

How does it work?

It uses word-embeddings generated by an AI model to represent words as vectors. Through training, the model learns to associate numbers with words in a way that words with similar meanings have vectors that are closer numerically. For example, "car" and "van" have similar vectors but "car" and "carpet" do not. The script then uses a similarity measure to compare the vector of the input word and the vector of the name of each country. The similarity score is then used to color the country on the map above.

Quirks

• Chad appears close to almost every word likely because it's a common English name.
  (Fixed by appending country names with "The country of" and embedding that string instead.)
• Similarly, Iraq appears close to almost every word probably because of abundance of WoT articles in the training data.

Specification

Model: text-embedding-3-small
Similarity measure:

• Cosine Similarity
• Euclidean Similarity

Tools:

• Plotly JS
• Vector operations in vanilla PHP

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.