search

Living-with-machines / DeezyMatch

A Flexible Deep Learning Approach to Fuzzy String Matching
https://living-with-machines.github.io/DeezyMatch/
Other
134 stars 34 forks source link

Active learning and fine-tuning #86

Open kasra-hosseini opened 3 years ago

kasra-hosseini commented 3 years ago

One possible (simple) implementation is to:

  1. Do model inference on a new dataset
  2. Based on the selected threshold (which can be a function argument), extract rows/examples in which DeezyMatch was most confused about. Create a new dataset for those rows.
  3. Currently, we don't have an annotation tool, but the user needs to annotate the extracted rows in step 2 for further fine-tuning.