search

GBADsInformatics / GBADsDataSearchUIAPI

FastAPI endpoint utilizing ML to recognize keywords and categories for GBADsDataSearchUI
GNU General Public License v3.0
0 stars 0 forks source link

Tied Words 🪢 #1

Closed MatthewSzurkowski closed 5 months ago

MatthewSzurkowski commented 11 months ago

Certain words need to be tied together like "North America" and "Polar bears" in a query.

kassyray commented 8 months ago

Are you thinking named entity disambiguation (NED) here?

MatthewSzurkowski commented 8 months ago

@kassyray thats exactly it! I didn't realize that there was a name for it, thank you.

MatthewSzurkowski commented 6 months ago

Almost complete, trying to employ a new NLP approach on a separate branch but will update main branch once all testing is complete.

MatthewSzurkowski commented 5 months ago

I've created functions to try and disambiguate entities for animal species, but the general results provide harsh returns and could potentially compromise this version of the API and return incorrect results. I've continued to search for a better solution by testing different NLP libraries, which could be promising, but the best bet is to use a custom trained model which will take some time. For version 1.0, I believe we should use what we have and let the GUI iron out any issues that arise. The GUI acts as a second form of authentication which allows the API results to have some give.

For future development, I'm leaving a note here for more packages to checkout:

I'll also be pushing the old commit to a separate branch called ned-requested-call