search

purplepotion / sadrat

Smart Adverse Drug Reaction Assessment Tools.
MIT License
17 stars 10 forks source link

Improve quality of Data Programming model #11

Open ShaswatLenka opened 4 years ago

ShaswatLenka commented 4 years ago

I have written a very basic Data Programming model for labeling the twitter dataset of adrmine. This currently is a "bad model" with an accuracy of 40% and biased to a single class in our binary classification among 0 and 1. This may have arisen due to a lot of reasons including -

  1. Quality of LFs(Labelling Functions).
  2. Coverage of quality LFs.
  3. Choice of ML model. I am self assigning this issue currently. But it would be great to have you work on this.
Debanitrkl commented 4 years ago

https://scispacy.apps.allenai.org/ check this: If we could get some help