feature importances with absolute values for decomposition-based methods

x-tabdeveloping / turftopic

Robust and fast topic models with sentence-transformers.

https://x-tabdeveloping.github.io/turftopic/

MIT License

8 stars 3 forks source link

feature importances with absolute values for decomposition-based methods #27

Closed rbroc closed 4 months ago

rbroc commented 4 months ago

just a pet project for myself to see how topics would look qualitatively if we selected keywords based on absolute (rather than strictly positive) values on the new dimensions

x-tabdeveloping commented 4 months ago

Coolio! I was actually also thinking that it might be really useful to add a possibility for also displaying negative terms in the descriptive tables when a model is already fitted.

Kinda like:

model.print_topics(display_negative=True)

Topic ID	Positive	Negative
0	Dawg, Darg, Hot Diggity Dog, ...	Cold Non-dog, Cool Catty-Cat, ...
1	Gooning, Edging, Henchmen, Spooning, ...	Girlbossing, Gatekeeping, CEO, ...

rbroc commented 4 months ago

closing because it overlaps with #29