Conversation

Replying to
i run the tweets through a transformer model to get semantic representations, then reduce dimensions to XY and run a clustering algorithm to find 'topics'. so each color is a different topic and similar tweets are near each other in XY space. the Z axis is time of course
3