Question-Classification-using-BERT

Question-Classification

Classifying questions from UIUC's CogComp QC Dataset

Classifying Questions into Coarse (6 classes) and Fine (50 classes) classes.

In here, we used pre-trained embeddings from BERT and applied transfer learning to see if the performance improves from the earlier experiments on the same dataset.

The code and explanation for the previous experimentation can be found at https://github.com/amankedia/Question-Classification.

The dataset can also be downloaded from the same repository.

Python and Tensforflow versions used

Python - 3.6

Tensorflow - 2.0

Results

Coarse Set Accuracy: 97.4%

Fine Set Accuracy: 92.2%

Accuracy Comparison with Previous Experimentation

Coarse Set Accuracy Improvement: 9.2%

Fine Set Accuracy Improvement: 10.6%

Ref: https://github.com/google-research/bert