As per suggested by @Jemoka:
Adding utilities for comparing topic models would make the library stand out from the rest of the crowd, and would make a more convincing point for adoption by most people.
I think the following would make quite a bit of sense:
- Embedding-based coherence on:
- Words
- Documents
- Images
- Diversity metrics:
- Embedding-based
- Lexical
It would also make a lot of sense to add the option for significance tests, since most of these metrics are aggregated.