Topics explorer / heatmap

Similarity map

Topics with similar text sit close together. Dot size shows hearts; colour shows the layer you pick.

Colour by

How similarity is worked out

  1. A small language model, run locally rather than through an online service () reads each topic's title and text and turns it into a list of 384 numbers that capture what it is about. Topics on similar subjects get similar numbers, even if they share no exact words.
  2. A method called t-SNE then arranges the topics on a flat map so that each one stays close to the topics whose numbers are most like its own. Dots that touched are nudged apart slightly.
  3. Themes and groups are assigned by comparing each topic's numbers with a short written description of every theme and group, and picking the closest. A few topics are placed by hand.

Dots that are near each other are similar. The axes have no meaning, and how far apart two distant clusters sit says little. Hearts, authors and dates play no part in where a dot sits.

Show all topics as a table