Closed dafajon closed 3 years ago
Current product review prebuilt model implementation suffers from low F1 score for public use. The dataset's NEUTRAL label is very noisy. In order make the PR appropriate for merge, I propose following approaches:
NEUTRAL
neutral
positive
negative
product_review
product_review_semisupervised
This will address pending request on #238
Addressed via a patch in e552b4b.
Current product review prebuilt model implementation suffers from low F1 score for public use. The dataset's
NEUTRAL
label is very noisy. In order make the PR appropriate for merge, I propose following approaches:neutral
label and re-label neutral instances aspositive
andnegative
with semi-supervised learning.product_review
andproduct_review_semisupervised
This will address pending request on #238