Hello .
actually, I want to perform text classification using Bert algorithm on a text data imbalanced . I have to balance my data but first convert text to vector and then balance my data. but I don't know how do this work as Bert has a special tokenizer.
thank you for your help
once you create batch of data, you have to pass it to bert tokenizer, then you can balance it via SMOTE, but i would say instead of doing this, augment your minority class
Dear Mahsa Mirzaei,
Interesting topic, so, in my opinion, BERT used, while capable of handling imbalanced classes with no additional data augmentation, does not generalize well when the training and test data are sufficiently dissimilar. Hence you first use Bert tokenizer then you can balance your data by using a sampling process, which is applied only to the training set and no changes are made to the validation and testing data. Imblearn library in python comes in handy to achieve the data resampling.
For Further Reference:
Conference Paper Automated Phrase Mining Using POST: The Best Approach
Conference Paper Odia Offline Typewritten Character Recognition using Templat...
Best wishes,
Jogeswar
- Badges
- Science topic
- Similar topics
- Vectorization