Latent Library

Inspired by this project, I took a day to spatially organize each sentence in The Great Gatsby by its embedding value in Google’s Universal Sentence Encoder. As such, related sentences are placed near each other, introducing a new way to conceptualize meaning. The website draws connections that might not be immediately visible on first read. While some are obvious (sentences about Jordan’s golf tournaments are all grouped together), others reveal less explicit undertones in the text (marriage and ownership are also identified as related concepts). 

Take a look here.


Using a raw .txt file from Project Gutenberg, I procsseced the each sentence in the book using Tensorflow. I reduced the the 512 output dimensions to three using and SKLearn’s TSNE and PCA algorithms. From there, I passed the data to a web app in JSON format. I used Three.js to represent the data visually. Check out the GitHub.

Find me on ︎︎, ︎, or ︎.