Emotion Atlas runs a real language model in the browser and maps thousands of emotion-labelled tweets in 3D.
The model, all-MiniLM-L6-v2, runs on the visitor’s graphics card through WebGPU. It turns each tweet from the public dair-ai/emotion dataset into a 384-dimensional vector that captures its meaning. The vectors are projected into 3D, so tweets with similar meaning sit close together, coloured by their emotion label.
Type a sentence and it is embedded live, placed on the map with lines to its nearest neighbours, and assigned an emotion by a nearest-neighbour vote. The model downloads once, about 4 MB, and is then cached.
Data: the public dair-ai/emotion dataset.
Lab demos are working experiments that show the methods behind our products. The same techniques can be built into software for your organisation.
See all demos →