Ok so I made some big changes. First of all, I updated the topic classification to a dedicated model, which I later hope to integrate with the visual representation. Additionally, I cleaned up code and explanations.

In the future I hope to:

Create a single python executable to run all the code and generation
Fully leaf topic generate on the research paper repository itself
Integrate new data sources
Improve data cleaning to recognize unknowns semantically
Also make topic representation fully semantic as well with a dot product distance.

Also, check out this topic hierarchy:!!!